diff --git a/ai/LICENSE.md b/ai/LICENSE.md new file mode 100644 index 0000000..f288702 --- /dev/null +++ b/ai/LICENSE.md @@ -0,0 +1,674 @@ + GNU GENERAL PUBLIC LICENSE + Version 3, 29 June 2007 + + Copyright (C) 2007 Free Software Foundation, Inc. + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + + Preamble + + The GNU General Public License is a free, copyleft license for +software and other kinds of works. + + The licenses for most software and other practical works are designed +to take away your freedom to share and change the works. By contrast, +the GNU General Public License is intended to guarantee your freedom to +share and change all versions of a program--to make sure it remains free +software for all its users. We, the Free Software Foundation, use the +GNU General Public License for most of our software; it applies also to +any other work released this way by its authors. You can apply it to +your programs, too. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +them if you wish), that you receive source code or can get it if you +want it, that you can change the software or use pieces of it in new +free programs, and that you know you can do these things. + + To protect your rights, we need to prevent others from denying you +these rights or asking you to surrender the rights. Therefore, you have +certain responsibilities if you distribute copies of the software, or if +you modify it: responsibilities to respect the freedom of others. + + For example, if you distribute copies of such a program, whether +gratis or for a fee, you must pass on to the recipients the same +freedoms that you received. You must make sure that they, too, receive +or can get the source code. And you must show them these terms so they +know their rights. + + Developers that use the GNU GPL protect your rights with two steps: +(1) assert copyright on the software, and (2) offer you this License +giving you legal permission to copy, distribute and/or modify it. + + For the developers' and authors' protection, the GPL clearly explains +that there is no warranty for this free software. For both users' and +authors' sake, the GPL requires that modified versions be marked as +changed, so that their problems will not be attributed erroneously to +authors of previous versions. + + Some devices are designed to deny users access to install or run +modified versions of the software inside them, although the manufacturer +can do so. This is fundamentally incompatible with the aim of +protecting users' freedom to change the software. The systematic +pattern of such abuse occurs in the area of products for individuals to +use, which is precisely where it is most unacceptable. Therefore, we +have designed this version of the GPL to prohibit the practice for those +products. If such problems arise substantially in other domains, we +stand ready to extend this provision to those domains in future versions +of the GPL, as needed to protect the freedom of users. + + Finally, every program is threatened constantly by software patents. +States should not allow patents to restrict development and use of +software on general-purpose computers, but in those that do, we wish to +avoid the special danger that patents applied to a free program could +make it effectively proprietary. To prevent this, the GPL assures that +patents cannot be used to render the program non-free. + + The precise terms and conditions for copying, distribution and +modification follow. + + TERMS AND CONDITIONS + + 0. Definitions. + + "This License" refers to version 3 of the GNU General Public License. + + "Copyright" also means copyright-like laws that apply to other kinds of +works, such as semiconductor masks. + + "The Program" refers to any copyrightable work licensed under this +License. Each licensee is addressed as "you". "Licensees" and +"recipients" may be individuals or organizations. + + To "modify" a work means to copy from or adapt all or part of the work +in a fashion requiring copyright permission, other than the making of an +exact copy. The resulting work is called a "modified version" of the +earlier work or a work "based on" the earlier work. + + A "covered work" means either the unmodified Program or a work based +on the Program. + + To "propagate" a work means to do anything with it that, without +permission, would make you directly or secondarily liable for +infringement under applicable copyright law, except executing it on a +computer or modifying a private copy. Propagation includes copying, +distribution (with or without modification), making available to the +public, and in some countries other activities as well. + + To "convey" a work means any kind of propagation that enables other +parties to make or receive copies. Mere interaction with a user through +a computer network, with no transfer of a copy, is not conveying. + + An interactive user interface displays "Appropriate Legal Notices" +to the extent that it includes a convenient and prominently visible +feature that (1) displays an appropriate copyright notice, and (2) +tells the user that there is no warranty for the work (except to the +extent that warranties are provided), that licensees may convey the +work under this License, and how to view a copy of this License. If +the interface presents a list of user commands or options, such as a +menu, a prominent item in the list meets this criterion. + + 1. Source Code. + + The "source code" for a work means the preferred form of the work +for making modifications to it. "Object code" means any non-source +form of a work. + + A "Standard Interface" means an interface that either is an official +standard defined by a recognized standards body, or, in the case of +interfaces specified for a particular programming language, one that +is widely used among developers working in that language. + + The "System Libraries" of an executable work include anything, other +than the work as a whole, that (a) is included in the normal form of +packaging a Major Component, but which is not part of that Major +Component, and (b) serves only to enable use of the work with that +Major Component, or to implement a Standard Interface for which an +implementation is available to the public in source code form. A +"Major Component", in this context, means a major essential component +(kernel, window system, and so on) of the specific operating system +(if any) on which the executable work runs, or a compiler used to +produce the work, or an object code interpreter used to run it. + + The "Corresponding Source" for a work in object code form means all +the source code needed to generate, install, and (for an executable +work) run the object code and to modify the work, including scripts to +control those activities. However, it does not include the work's +System Libraries, or general-purpose tools or generally available free +programs which are used unmodified in performing those activities but +which are not part of the work. For example, Corresponding Source +includes interface definition files associated with source files for +the work, and the source code for shared libraries and dynamically +linked subprograms that the work is specifically designed to require, +such as by intimate data communication or control flow between those +subprograms and other parts of the work. + + The Corresponding Source need not include anything that users +can regenerate automatically from other parts of the Corresponding +Source. + + The Corresponding Source for a work in source code form is that +same work. + + 2. Basic Permissions. + + All rights granted under this License are granted for the term of +copyright on the Program, and are irrevocable provided the stated +conditions are met. This License explicitly affirms your unlimited +permission to run the unmodified Program. The output from running a +covered work is covered by this License only if the output, given its +content, constitutes a covered work. This License acknowledges your +rights of fair use or other equivalent, as provided by copyright law. + + You may make, run and propagate covered works that you do not +convey, without conditions so long as your license otherwise remains +in force. You may convey covered works to others for the sole purpose +of having them make modifications exclusively for you, or provide you +with facilities for running those works, provided that you comply with +the terms of this License in conveying all material for which you do +not control copyright. Those thus making or running the covered works +for you must do so exclusively on your behalf, under your direction +and control, on terms that prohibit them from making any copies of +your copyrighted material outside their relationship with you. + + Conveying under any other circumstances is permitted solely under +the conditions stated below. Sublicensing is not allowed; section 10 +makes it unnecessary. + + 3. Protecting Users' Legal Rights From Anti-Circumvention Law. + + No covered work shall be deemed part of an effective technological +measure under any applicable law fulfilling obligations under article +11 of the WIPO copyright treaty adopted on 20 December 1996, or +similar laws prohibiting or restricting circumvention of such +measures. + + When you convey a covered work, you waive any legal power to forbid +circumvention of technological measures to the extent such circumvention +is effected by exercising rights under this License with respect to +the covered work, and you disclaim any intention to limit operation or +modification of the work as a means of enforcing, against the work's +users, your or third parties' legal rights to forbid circumvention of +technological measures. + + 4. Conveying Verbatim Copies. + + You may convey verbatim copies of the Program's source code as you +receive it, in any medium, provided that you conspicuously and +appropriately publish on each copy an appropriate copyright notice; +keep intact all notices stating that this License and any +non-permissive terms added in accord with section 7 apply to the code; +keep intact all notices of the absence of any warranty; and give all +recipients a copy of this License along with the Program. + + You may charge any price or no price for each copy that you convey, +and you may offer support or warranty protection for a fee. + + 5. Conveying Modified Source Versions. + + You may convey a work based on the Program, or the modifications to +produce it from the Program, in the form of source code under the +terms of section 4, provided that you also meet all of these conditions: + + a) The work must carry prominent notices stating that you modified + it, and giving a relevant date. + + b) The work must carry prominent notices stating that it is + released under this License and any conditions added under section + 7. This requirement modifies the requirement in section 4 to + "keep intact all notices". + + c) You must license the entire work, as a whole, under this + License to anyone who comes into possession of a copy. This + License will therefore apply, along with any applicable section 7 + additional terms, to the whole of the work, and all its parts, + regardless of how they are packaged. This License gives no + permission to license the work in any other way, but it does not + invalidate such permission if you have separately received it. + + d) If the work has interactive user interfaces, each must display + Appropriate Legal Notices; however, if the Program has interactive + interfaces that do not display Appropriate Legal Notices, your + work need not make them do so. + + A compilation of a covered work with other separate and independent +works, which are not by their nature extensions of the covered work, +and which are not combined with it such as to form a larger program, +in or on a volume of a storage or distribution medium, is called an +"aggregate" if the compilation and its resulting copyright are not +used to limit the access or legal rights of the compilation's users +beyond what the individual works permit. Inclusion of a covered work +in an aggregate does not cause this License to apply to the other +parts of the aggregate. + + 6. Conveying Non-Source Forms. + + You may convey a covered work in object code form under the terms +of sections 4 and 5, provided that you also convey the +machine-readable Corresponding Source under the terms of this License, +in one of these ways: + + a) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by the + Corresponding Source fixed on a durable physical medium + customarily used for software interchange. + + b) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by a + written offer, valid for at least three years and valid for as + long as you offer spare parts or customer support for that product + model, to give anyone who possesses the object code either (1) a + copy of the Corresponding Source for all the software in the + product that is covered by this License, on a durable physical + medium customarily used for software interchange, for a price no + more than your reasonable cost of physically performing this + conveying of source, or (2) access to copy the + Corresponding Source from a network server at no charge. + + c) Convey individual copies of the object code with a copy of the + written offer to provide the Corresponding Source. This + alternative is allowed only occasionally and noncommercially, and + only if you received the object code with such an offer, in accord + with subsection 6b. + + d) Convey the object code by offering access from a designated + place (gratis or for a charge), and offer equivalent access to the + Corresponding Source in the same way through the same place at no + further charge. You need not require recipients to copy the + Corresponding Source along with the object code. If the place to + copy the object code is a network server, the Corresponding Source + may be on a different server (operated by you or a third party) + that supports equivalent copying facilities, provided you maintain + clear directions next to the object code saying where to find the + Corresponding Source. Regardless of what server hosts the + Corresponding Source, you remain obligated to ensure that it is + available for as long as needed to satisfy these requirements. + + e) Convey the object code using peer-to-peer transmission, provided + you inform other peers where the object code and Corresponding + Source of the work are being offered to the general public at no + charge under subsection 6d. + + A separable portion of the object code, whose source code is excluded +from the Corresponding Source as a System Library, need not be +included in conveying the object code work. + + A "User Product" is either (1) a "consumer product", which means any +tangible personal property which is normally used for personal, family, +or household purposes, or (2) anything designed or sold for incorporation +into a dwelling. In determining whether a product is a consumer product, +doubtful cases shall be resolved in favor of coverage. For a particular +product received by a particular user, "normally used" refers to a +typical or common use of that class of product, regardless of the status +of the particular user or of the way in which the particular user +actually uses, or expects or is expected to use, the product. A product +is a consumer product regardless of whether the product has substantial +commercial, industrial or non-consumer uses, unless such uses represent +the only significant mode of use of the product. + + "Installation Information" for a User Product means any methods, +procedures, authorization keys, or other information required to install +and execute modified versions of a covered work in that User Product from +a modified version of its Corresponding Source. The information must +suffice to ensure that the continued functioning of the modified object +code is in no case prevented or interfered with solely because +modification has been made. + + If you convey an object code work under this section in, or with, or +specifically for use in, a User Product, and the conveying occurs as +part of a transaction in which the right of possession and use of the +User Product is transferred to the recipient in perpetuity or for a +fixed term (regardless of how the transaction is characterized), the +Corresponding Source conveyed under this section must be accompanied +by the Installation Information. But this requirement does not apply +if neither you nor any third party retains the ability to install +modified object code on the User Product (for example, the work has +been installed in ROM). + + The requirement to provide Installation Information does not include a +requirement to continue to provide support service, warranty, or updates +for a work that has been modified or installed by the recipient, or for +the User Product in which it has been modified or installed. Access to a +network may be denied when the modification itself materially and +adversely affects the operation of the network or violates the rules and +protocols for communication across the network. + + Corresponding Source conveyed, and Installation Information provided, +in accord with this section must be in a format that is publicly +documented (and with an implementation available to the public in +source code form), and must require no special password or key for +unpacking, reading or copying. + + 7. Additional Terms. + + "Additional permissions" are terms that supplement the terms of this +License by making exceptions from one or more of its conditions. +Additional permissions that are applicable to the entire Program shall +be treated as though they were included in this License, to the extent +that they are valid under applicable law. If additional permissions +apply only to part of the Program, that part may be used separately +under those permissions, but the entire Program remains governed by +this License without regard to the additional permissions. + + When you convey a copy of a covered work, you may at your option +remove any additional permissions from that copy, or from any part of +it. (Additional permissions may be written to require their own +removal in certain cases when you modify the work.) You may place +additional permissions on material, added by you to a covered work, +for which you have or can give appropriate copyright permission. + + Notwithstanding any other provision of this License, for material you +add to a covered work, you may (if authorized by the copyright holders of +that material) supplement the terms of this License with terms: + + a) Disclaiming warranty or limiting liability differently from the + terms of sections 15 and 16 of this License; or + + b) Requiring preservation of specified reasonable legal notices or + author attributions in that material or in the Appropriate Legal + Notices displayed by works containing it; or + + c) Prohibiting misrepresentation of the origin of that material, or + requiring that modified versions of such material be marked in + reasonable ways as different from the original version; or + + d) Limiting the use for publicity purposes of names of licensors or + authors of the material; or + + e) Declining to grant rights under trademark law for use of some + trade names, trademarks, or service marks; or + + f) Requiring indemnification of licensors and authors of that + material by anyone who conveys the material (or modified versions of + it) with contractual assumptions of liability to the recipient, for + any liability that these contractual assumptions directly impose on + those licensors and authors. + + All other non-permissive additional terms are considered "further +restrictions" within the meaning of section 10. If the Program as you +received it, or any part of it, contains a notice stating that it is +governed by this License along with a term that is a further +restriction, you may remove that term. If a license document contains +a further restriction but permits relicensing or conveying under this +License, you may add to a covered work material governed by the terms +of that license document, provided that the further restriction does +not survive such relicensing or conveying. + + If you add terms to a covered work in accord with this section, you +must place, in the relevant source files, a statement of the +additional terms that apply to those files, or a notice indicating +where to find the applicable terms. + + Additional terms, permissive or non-permissive, may be stated in the +form of a separately written license, or stated as exceptions; +the above requirements apply either way. + + 8. Termination. + + You may not propagate or modify a covered work except as expressly +provided under this License. Any attempt otherwise to propagate or +modify it is void, and will automatically terminate your rights under +this License (including any patent licenses granted under the third +paragraph of section 11). + + However, if you cease all violation of this License, then your +license from a particular copyright holder is reinstated (a) +provisionally, unless and until the copyright holder explicitly and +finally terminates your license, and (b) permanently, if the copyright +holder fails to notify you of the violation by some reasonable means +prior to 60 days after the cessation. + + Moreover, your license from a particular copyright holder is +reinstated permanently if the copyright holder notifies you of the +violation by some reasonable means, this is the first time you have +received notice of violation of this License (for any work) from that +copyright holder, and you cure the violation prior to 30 days after +your receipt of the notice. + + Termination of your rights under this section does not terminate the +licenses of parties who have received copies or rights from you under +this License. If your rights have been terminated and not permanently +reinstated, you do not qualify to receive new licenses for the same +material under section 10. + + 9. Acceptance Not Required for Having Copies. + + You are not required to accept this License in order to receive or +run a copy of the Program. Ancillary propagation of a covered work +occurring solely as a consequence of using peer-to-peer transmission +to receive a copy likewise does not require acceptance. However, +nothing other than this License grants you permission to propagate or +modify any covered work. These actions infringe copyright if you do +not accept this License. Therefore, by modifying or propagating a +covered work, you indicate your acceptance of this License to do so. + + 10. Automatic Licensing of Downstream Recipients. + + Each time you convey a covered work, the recipient automatically +receives a license from the original licensors, to run, modify and +propagate that work, subject to this License. You are not responsible +for enforcing compliance by third parties with this License. + + An "entity transaction" is a transaction transferring control of an +organization, or substantially all assets of one, or subdividing an +organization, or merging organizations. If propagation of a covered +work results from an entity transaction, each party to that +transaction who receives a copy of the work also receives whatever +licenses to the work the party's predecessor in interest had or could +give under the previous paragraph, plus a right to possession of the +Corresponding Source of the work from the predecessor in interest, if +the predecessor has it or can get it with reasonable efforts. + + You may not impose any further restrictions on the exercise of the +rights granted or affirmed under this License. For example, you may +not impose a license fee, royalty, or other charge for exercise of +rights granted under this License, and you may not initiate litigation +(including a cross-claim or counterclaim in a lawsuit) alleging that +any patent claim is infringed by making, using, selling, offering for +sale, or importing the Program or any portion of it. + + 11. Patents. + + A "contributor" is a copyright holder who authorizes use under this +License of the Program or a work on which the Program is based. The +work thus licensed is called the contributor's "contributor version". + + A contributor's "essential patent claims" are all patent claims +owned or controlled by the contributor, whether already acquired or +hereafter acquired, that would be infringed by some manner, permitted +by this License, of making, using, or selling its contributor version, +but do not include claims that would be infringed only as a +consequence of further modification of the contributor version. For +purposes of this definition, "control" includes the right to grant +patent sublicenses in a manner consistent with the requirements of +this License. + + Each contributor grants you a non-exclusive, worldwide, royalty-free +patent license under the contributor's essential patent claims, to +make, use, sell, offer for sale, import and otherwise run, modify and +propagate the contents of its contributor version. + + In the following three paragraphs, a "patent license" is any express +agreement or commitment, however denominated, not to enforce a patent +(such as an express permission to practice a patent or covenant not to +sue for patent infringement). To "grant" such a patent license to a +party means to make such an agreement or commitment not to enforce a +patent against the party. + + If you convey a covered work, knowingly relying on a patent license, +and the Corresponding Source of the work is not available for anyone +to copy, free of charge and under the terms of this License, through a +publicly available network server or other readily accessible means, +then you must either (1) cause the Corresponding Source to be so +available, or (2) arrange to deprive yourself of the benefit of the +patent license for this particular work, or (3) arrange, in a manner +consistent with the requirements of this License, to extend the patent +license to downstream recipients. "Knowingly relying" means you have +actual knowledge that, but for the patent license, your conveying the +covered work in a country, or your recipient's use of the covered work +in a country, would infringe one or more identifiable patents in that +country that you have reason to believe are valid. + + If, pursuant to or in connection with a single transaction or +arrangement, you convey, or propagate by procuring conveyance of, a +covered work, and grant a patent license to some of the parties +receiving the covered work authorizing them to use, propagate, modify +or convey a specific copy of the covered work, then the patent license +you grant is automatically extended to all recipients of the covered +work and works based on it. + + A patent license is "discriminatory" if it does not include within +the scope of its coverage, prohibits the exercise of, or is +conditioned on the non-exercise of one or more of the rights that are +specifically granted under this License. You may not convey a covered +work if you are a party to an arrangement with a third party that is +in the business of distributing software, under which you make payment +to the third party based on the extent of your activity of conveying +the work, and under which the third party grants, to any of the +parties who would receive the covered work from you, a discriminatory +patent license (a) in connection with copies of the covered work +conveyed by you (or copies made from those copies), or (b) primarily +for and in connection with specific products or compilations that +contain the covered work, unless you entered into that arrangement, +or that patent license was granted, prior to 28 March 2007. + + Nothing in this License shall be construed as excluding or limiting +any implied license or other defenses to infringement that may +otherwise be available to you under applicable patent law. + + 12. No Surrender of Others' Freedom. + + If conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot convey a +covered work so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you may +not convey it at all. For example, if you agree to terms that obligate you +to collect a royalty for further conveying from those to whom you convey +the Program, the only way you could satisfy both those terms and this +License would be to refrain entirely from conveying the Program. + + 13. Use with the GNU Affero General Public License. + + Notwithstanding any other provision of this License, you have +permission to link or combine any covered work with a work licensed +under version 3 of the GNU Affero General Public License into a single +combined work, and to convey the resulting work. The terms of this +License will continue to apply to the part which is the covered work, +but the special requirements of the GNU Affero General Public License, +section 13, concerning interaction through a network will apply to the +combination as such. + + 14. Revised Versions of this License. + + The Free Software Foundation may publish revised and/or new versions of +the GNU General Public License from time to time. Such new versions will +be similar in spirit to the present version, but may differ in detail to +address new problems or concerns. + + Each version is given a distinguishing version number. If the +Program specifies that a certain numbered version of the GNU General +Public License "or any later version" applies to it, you have the +option of following the terms and conditions either of that numbered +version or of any later version published by the Free Software +Foundation. If the Program does not specify a version number of the +GNU General Public License, you may choose any version ever published +by the Free Software Foundation. + + If the Program specifies that a proxy can decide which future +versions of the GNU General Public License can be used, that proxy's +public statement of acceptance of a version permanently authorizes you +to choose that version for the Program. + + Later license versions may give you additional or different +permissions. However, no additional obligations are imposed on any +author or copyright holder as a result of your choosing to follow a +later version. + + 15. Disclaimer of Warranty. + + THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY +APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT +HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY +OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, +THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM +IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF +ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. Limitation of Liability. + + IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING +WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS +THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY +GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE +USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF +DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD +PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), +EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF +SUCH DAMAGES. + + 17. Interpretation of Sections 15 and 16. + + If the disclaimer of warranty and limitation of liability provided +above cannot be given local legal effect according to their terms, +reviewing courts shall apply local law that most closely approximates +an absolute waiver of all civil liability in connection with the +Program, unless a warranty or assumption of liability accompanies a +copy of the Program in return for a fee. + + END OF TERMS AND CONDITIONS + + How to Apply These Terms to Your New Programs + + If you develop a new program, and you want it to be of the greatest +possible use to the public, the best way to achieve this is to make it +free software which everyone can redistribute and change under these terms. + + To do so, attach the following notices to the program. It is safest +to attach them to the start of each source file to most effectively +state the exclusion of warranty; and each file should have at least +the "copyright" line and a pointer to where the full notice is found. + + + Copyright (C) + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + +Also add information on how to contact you by electronic and paper mail. + + If the program does terminal interaction, make it output a short +notice like this when it starts in an interactive mode: + + Copyright (C) + This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. + This is free software, and you are welcome to redistribute it + under certain conditions; type `show c' for details. + +The hypothetical commands `show w' and `show c' should show the appropriate +parts of the General Public License. Of course, your program's commands +might be different; for a GUI interface, you would use an "about box". + + You should also get your employer (if you work as a programmer) or school, +if any, to sign a "copyright disclaimer" for the program, if necessary. +For more information on this, and how to apply and follow the GNU GPL, see +. + + The GNU General Public License does not permit incorporating your program +into proprietary programs. If your program is a subroutine library, you +may consider it more useful to permit linking proprietary applications with +the library. If this is what you want to do, use the GNU Lesser General +Public License instead of this License. But first, please read +. diff --git a/ai/README.md b/ai/README.md new file mode 100644 index 0000000..6ed745d --- /dev/null +++ b/ai/README.md @@ -0,0 +1,29 @@ +# Training the PyTorch Localizer and Detector + +## Content + +The `model.py` files contains a Python class called `Dataset`, which has methods for manipulating datasets of SkyScan images and training models on them. There are also some example functions that use the class, including `revised_sequence()`, which runs an entire data analysis pipeline. The steps in that pipeline including ingesting the imagery, training a localizer (which finds bounding boxes), using the localizer to find bounding boxes on additional images, training a detector (which finds bounding boxes and aircraft class), and testing the performance of localizer and detector with testing data. + +The pipeline will need to be adapted to meet your specific needs. Each time a model is trained or used, YOLOv7 outputs the results to a folder. To retrain a model or rerun a test, the corresponding folder name should be changed in the code to avoid a collision. On the other hand, if you've already trained a model or run a test and are rerunning the code, you can instead comment out the corresponding `.train()` and `.test()` calls to save time. In general, lines ending with `##` in `revised_sequence()` are time-intensive steps that only need to be run once, and can be commented out to save time if rerunning the code. + +## Setup + +Install the [YOLOv7 repo](https://github.com/WongKinYiu/yolov7), following the instructions there. Then, copy the following files into the root folder of your YOLOv7 installation: +* `model.py` +* `test2.py` +* `localizer_rev2.yaml` +* `detector_rev2.yaml` +* `taxon.json` + +Copy the following files into the `utils` folder: +* `metrics2.py` +* `datasets2.py` + +Then, edit the `model.py` file's `revised_sequence()` method and update the file paths to point to the appropriate folders for your system. File paths in `localizer_rev2.yaml` and `detector_rev2.yaml` will also need to be updated. + +Finally, from the YOLOv7 root folder, run: +`./model.py` + +## License + +The contents of this folder are derived from https://github.com/WongKinYiu/yolov7 and are being released under the GNU General Public License v3.0. (The remainder of this repo is released under the Apache License 2.0.) diff --git a/ai/datasets2.py b/ai/datasets2.py new file mode 100644 index 0000000..44e14e2 --- /dev/null +++ b/ai/datasets2.py @@ -0,0 +1,1354 @@ +# Dataset utils and dataloaders + +import glob +import logging +import math +import os +import random +import shutil +import time +from itertools import repeat +from multiprocessing.pool import ThreadPool +from pathlib import Path +from threading import Thread + +import cv2 +import numpy as np +import torch +import torch.nn.functional as F +from PIL import Image, ExifTags +from torch.utils.data import Dataset +from tqdm import tqdm + +import pickle +from copy import deepcopy +#from pycocotools import mask as maskUtils +from torchvision.utils import save_image +from torchvision.ops import roi_pool, roi_align, ps_roi_pool, ps_roi_align + +from utils.general import check_requirements, xyxy2xywh, xywh2xyxy, xywhn2xyxy, xyn2xy, segment2box, segments2boxes, \ + resample_segments, clean_str +from utils.torch_utils import torch_distributed_zero_first + +# Parameters +help_url = 'https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data' +img_formats = ['bmp', 'jpg', 'jpeg', 'png', 'tif', 'tiff', 'dng', 'webp', 'mpo'] # acceptable image suffixes +vid_formats = ['mov', 'avi', 'mp4', 'mpg', 'mpeg', 'm4v', 'wmv', 'mkv'] # acceptable video suffixes +logger = logging.getLogger(__name__) + +# Get orientation exif tag +for orientation in ExifTags.TAGS.keys(): + if ExifTags.TAGS[orientation] == 'Orientation': + break + + +def get_hash(files): + # Returns a single hash value of a list of files + return sum(os.path.getsize(f) for f in files if os.path.isfile(f)) + + +def exif_size(img): + # Returns exif-corrected PIL size + s = img.size # (width, height) + try: + rotation = dict(img._getexif().items())[orientation] + if rotation == 6: # rotation 270 + s = (s[1], s[0]) + elif rotation == 8: # rotation 90 + s = (s[1], s[0]) + except: + pass + + return s + + +def create_dataloader(path, imgsz, batch_size, stride, opt, hyp=None, augment=False, cache=False, pad=0.0, rect=False, + rank=-1, world_size=1, workers=8, image_weights=False, quad=False, prefix=''): + # Make sure only the first process in DDP process the dataset first, and the following others can use the cache + with torch_distributed_zero_first(rank): + dataset = LoadImagesAndLabels(path, imgsz, batch_size, + augment=augment, # augment images + hyp=hyp, # augmentation hyperparameters + rect=rect, # rectangular training + cache_images=cache, + single_cls=opt.single_cls, + stride=int(stride), + pad=pad, + image_weights=image_weights, + prefix=prefix) + + batch_size = min(batch_size, len(dataset)) + nw = min([os.cpu_count() // world_size, batch_size if batch_size > 1 else 0, workers]) # number of workers + sampler = torch.utils.data.distributed.DistributedSampler(dataset) if rank != -1 else None + loader = torch.utils.data.DataLoader if image_weights else InfiniteDataLoader + # Use torch.utils.data.DataLoader() if dataset.properties will update during training else InfiniteDataLoader() + dataloader = loader(dataset, + batch_size=batch_size, + num_workers=nw, + sampler=sampler, + pin_memory=True, + collate_fn=LoadImagesAndLabels.collate_fn4 if quad else LoadImagesAndLabels.collate_fn) + return dataloader, dataset + + +class InfiniteDataLoader(torch.utils.data.dataloader.DataLoader): + """ Dataloader that reuses workers + + Uses same syntax as vanilla DataLoader + """ + + def __init__(self, *args, **kwargs): + super().__init__(*args, **kwargs) + object.__setattr__(self, 'batch_sampler', _RepeatSampler(self.batch_sampler)) + self.iterator = super().__iter__() + + def __len__(self): + return len(self.batch_sampler.sampler) + + def __iter__(self): + for i in range(len(self)): + yield next(self.iterator) + + +class _RepeatSampler(object): + """ Sampler that repeats forever + + Args: + sampler (Sampler) + """ + + def __init__(self, sampler): + self.sampler = sampler + + def __iter__(self): + while True: + yield from iter(self.sampler) + + +class LoadImages: # for inference + def __init__(self, path, img_size=640, stride=32): + p = str(Path(path).absolute()) # os-agnostic absolute path + if '*' in p: + files = sorted(glob.glob(p, recursive=True)) # glob + elif os.path.isdir(p): + files = sorted(glob.glob(os.path.join(p, '*.*'))) # dir + elif os.path.isfile(p): + files = [p] # files + else: + raise Exception(f'ERROR: {p} does not exist') + + images = [x for x in files if x.split('.')[-1].lower() in img_formats] + videos = [x for x in files if x.split('.')[-1].lower() in vid_formats] + ni, nv = len(images), len(videos) + + self.img_size = img_size + self.stride = stride + self.files = images + videos + self.nf = ni + nv # number of files + self.video_flag = [False] * ni + [True] * nv + self.mode = 'image' + if any(videos): + self.new_video(videos[0]) # new video + else: + self.cap = None + assert self.nf > 0, f'No images or videos found in {p}. ' \ + f'Supported formats are:\nimages: {img_formats}\nvideos: {vid_formats}' + + def __iter__(self): + self.count = 0 + return self + + def __next__(self): + if self.count == self.nf: + raise StopIteration + path = self.files[self.count] + + if self.video_flag[self.count]: + # Read video + self.mode = 'video' + ret_val, img0 = self.cap.read() + if not ret_val: + self.count += 1 + self.cap.release() + if self.count == self.nf: # last video + raise StopIteration + else: + path = self.files[self.count] + self.new_video(path) + ret_val, img0 = self.cap.read() + + self.frame += 1 + print(f'video {self.count + 1}/{self.nf} ({self.frame}/{self.nframes}) {path}: ', end='') + + else: + # Read image + self.count += 1 + img0 = cv2.imread(path) # BGR + assert img0 is not None, 'Image Not Found ' + path + #print(f'image {self.count}/{self.nf} {path}: ', end='') + + # Padded resize + img = letterbox(img0, self.img_size, stride=self.stride)[0] + + # Convert + img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416 + img = np.ascontiguousarray(img) + + return path, img, img0, self.cap + + def new_video(self, path): + self.frame = 0 + self.cap = cv2.VideoCapture(path) + self.nframes = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT)) + + def __len__(self): + return self.nf # number of files + + +class LoadWebcam: # for inference + def __init__(self, pipe='0', img_size=640, stride=32): + self.img_size = img_size + self.stride = stride + + if pipe.isnumeric(): + pipe = eval(pipe) # local camera + # pipe = 'rtsp://192.168.1.64/1' # IP camera + # pipe = 'rtsp://username:password@192.168.1.64/1' # IP camera with login + # pipe = 'http://wmccpinetop.axiscam.net/mjpg/video.mjpg' # IP golf camera + + self.pipe = pipe + self.cap = cv2.VideoCapture(pipe) # video capture object + self.cap.set(cv2.CAP_PROP_BUFFERSIZE, 3) # set buffer size + + def __iter__(self): + self.count = -1 + return self + + def __next__(self): + self.count += 1 + if cv2.waitKey(1) == ord('q'): # q to quit + self.cap.release() + cv2.destroyAllWindows() + raise StopIteration + + # Read frame + if self.pipe == 0: # local camera + ret_val, img0 = self.cap.read() + img0 = cv2.flip(img0, 1) # flip left-right + else: # IP camera + n = 0 + while True: + n += 1 + self.cap.grab() + if n % 30 == 0: # skip frames + ret_val, img0 = self.cap.retrieve() + if ret_val: + break + + # Print + assert ret_val, f'Camera Error {self.pipe}' + img_path = 'webcam.jpg' + print(f'webcam {self.count}: ', end='') + + # Padded resize + img = letterbox(img0, self.img_size, stride=self.stride)[0] + + # Convert + img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416 + img = np.ascontiguousarray(img) + + return img_path, img, img0, None + + def __len__(self): + return 0 + + +class LoadFromDir: # for inference + def __init__(self, path, img_size=640, stride=32): + self.path = path + self.img_size = img_size + self.stride = stride + self.mode = 'image' + + def __iter__(self): + self.count = 0 + return self + + def __next__(self): + # Check for files in folder, waiting if necessary + while True: + if os.path.isdir(self.path): + img_paths = glob.glob(os.path.join(self.path, '*')) + if len(img_paths) > 0: + break + time.sleep(0.5) + img_path = min(img_paths, key=os.path.getctime) + + # Process image + self.count += 1 + img0 = cv2.imread(img_path) + assert img0 is not None, 'Image Not Found ' + img_path + img = letterbox(img0, self.img_size, stride=self.stride)[0] # Size + img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, channel first + img = np.ascontiguousarray(img) + return img_path, img, img0, None + + def __len__(self): + return 0 + + +class LoadStreams: # multiple IP or RTSP cameras + def __init__(self, sources='streams.txt', img_size=640, stride=32): + self.mode = 'stream' + self.img_size = img_size + self.stride = stride + + if os.path.isfile(sources): + with open(sources, 'r') as f: + sources = [x.strip() for x in f.read().strip().splitlines() if len(x.strip())] + else: + sources = [sources] + + n = len(sources) + self.imgs = [None] * n + self.sources = [clean_str(x) for x in sources] # clean source names for later + for i, s in enumerate(sources): + # Start the thread to read frames from the video stream + print(f'{i + 1}/{n}: {s}... ', end='') + url = eval(s) if s.isnumeric() else s + if 'youtube.com/' in str(url) or 'youtu.be/' in str(url): # if source is YouTube video + check_requirements(('pafy', 'youtube_dl')) + import pafy + url = pafy.new(url).getbest(preftype="mp4").url + cap = cv2.VideoCapture(url) + assert cap.isOpened(), f'Failed to open {s}' + w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) + h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) + self.fps = cap.get(cv2.CAP_PROP_FPS) % 100 + + _, self.imgs[i] = cap.read() # guarantee first frame + thread = Thread(target=self.update, args=([i, cap]), daemon=True) + print(f' success ({w}x{h} at {self.fps:.2f} FPS).') + thread.start() + print('') # newline + + # check for common shapes + s = np.stack([letterbox(x, self.img_size, stride=self.stride)[0].shape for x in self.imgs], 0) # shapes + self.rect = np.unique(s, axis=0).shape[0] == 1 # rect inference if all shapes equal + if not self.rect: + print('WARNING: Different stream shapes detected. For optimal performance supply similarly-shaped streams.') + + def update(self, index, cap): + # Read next stream frame in a daemon thread + n = 0 + while cap.isOpened(): + n += 1 + # _, self.imgs[index] = cap.read() + cap.grab() + if n == 4: # read every 4th frame + success, im = cap.retrieve() + self.imgs[index] = im if success else self.imgs[index] * 0 + n = 0 + time.sleep(1 / self.fps) # wait time + + def __iter__(self): + self.count = -1 + return self + + def __next__(self): + self.count += 1 + img0 = self.imgs.copy() + if cv2.waitKey(1) == ord('q'): # q to quit + cv2.destroyAllWindows() + raise StopIteration + + # Letterbox + img = [letterbox(x, self.img_size, auto=self.rect, stride=self.stride)[0] for x in img0] + + # Stack + img = np.stack(img, 0) + + # Convert + img = img[:, :, :, ::-1].transpose(0, 3, 1, 2) # BGR to RGB, to bsx3x416x416 + img = np.ascontiguousarray(img) + + return self.sources, img, img0, None + + def __len__(self): + return 0 # 1E12 frames = 32 streams at 30 FPS for 30 years + + +def img2label_paths(img_paths): + # Define label paths as a function of image paths + sa, sb = os.sep + 'images' + os.sep, os.sep + 'labels' + os.sep # /images/, /labels/ substrings + return ['txt'.join(x.replace(sa, sb, 1).rsplit(x.split('.')[-1], 1)) for x in img_paths] + + +class LoadImagesAndLabels(Dataset): # for training/testing + def __init__(self, path, img_size=640, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False, + cache_images=False, single_cls=False, stride=32, pad=0.0, prefix=''): + self.img_size = img_size + self.augment = augment + self.hyp = hyp + self.image_weights = image_weights + self.rect = False if image_weights else rect + self.mosaic = self.augment and not self.rect # load 4 images at a time into a mosaic (only during training) + self.mosaic_border = [-img_size // 2, -img_size // 2] + self.stride = stride + self.path = path + #self.albumentations = Albumentations() if augment else None + + try: + f = [] # image files + for p in path if isinstance(path, list) else [path]: + p = Path(p) # os-agnostic + if p.is_dir(): # dir + f += glob.glob(str(p / '**' / '*.*'), recursive=True) + # f = list(p.rglob('**/*.*')) # pathlib + elif p.is_file(): # file + with open(p, 'r') as t: + t = t.read().strip().splitlines() + parent = str(p.parent) + os.sep + f += [x.replace('./', parent) if x.startswith('./') else x for x in t] # local to global path + # f += [p.parent / x.lstrip(os.sep) for x in t] # local to global path (pathlib) + else: + raise Exception(f'{prefix}{p} does not exist') + self.img_files = sorted([x.replace('/', os.sep) for x in f if x.split('.')[-1].lower() in img_formats]) + # self.img_files = sorted([x for x in f if x.suffix[1:].lower() in img_formats]) # pathlib + assert self.img_files, f'{prefix}No images found' + except Exception as e: + raise Exception(f'{prefix}Error loading data from {path}: {e}\nSee {help_url}') + + # Check cache + self.label_files = img2label_paths(self.img_files) # labels + cache_path = (p if p.is_file() else Path(self.label_files[0]).parent).with_suffix('.cache') # cached labels + if cache_path.is_file(): + cache, exists = torch.load(cache_path), True # load + #if cache['hash'] != get_hash(self.label_files + self.img_files) or 'version' not in cache: # changed + # cache, exists = self.cache_labels(cache_path, prefix), False # re-cache + else: + cache, exists = self.cache_labels(cache_path, prefix), False # cache + + # Display cache + nf, nm, ne, nc, n = cache.pop('results') # found, missing, empty, corrupted, total + if exists: + d = f"Scanning '{cache_path}' images and labels... {nf} found, {nm} missing, {ne} empty, {nc} corrupted" + tqdm(None, desc=prefix + d, total=n, initial=n) # display cache results + assert nf > 0 or not augment, f'{prefix}No labels in {cache_path}. Can not train without labels. See {help_url}' + + # Read cache + cache.pop('hash') # remove hash + cache.pop('version') # remove version + labels, shapes, self.segments = zip(*cache.values()) + self.labels = list(labels) + self.shapes = np.array(shapes, dtype=np.float64) + self.img_files = list(cache.keys()) # update + self.label_files = img2label_paths(cache.keys()) # update + if single_cls: + for x in self.labels: + x[:, 0] = 0 + + n = len(shapes) # number of images + bi = np.floor(np.arange(n) / batch_size).astype(np.int) # batch index + nb = bi[-1] + 1 # number of batches + self.batch = bi # batch index of image + self.n = n + self.indices = range(n) + + # Rectangular Training + if self.rect: + # Sort by aspect ratio + s = self.shapes # wh + ar = s[:, 1] / s[:, 0] # aspect ratio + irect = ar.argsort() + self.img_files = [self.img_files[i] for i in irect] + self.label_files = [self.label_files[i] for i in irect] + self.labels = [self.labels[i] for i in irect] + self.shapes = s[irect] # wh + ar = ar[irect] + + # Set training image shapes + shapes = [[1, 1]] * nb + for i in range(nb): + ari = ar[bi == i] + mini, maxi = ari.min(), ari.max() + if maxi < 1: + shapes[i] = [maxi, 1] + elif mini > 1: + shapes[i] = [1, 1 / mini] + + self.batch_shapes = np.ceil(np.array(shapes) * img_size / stride + pad).astype(np.int) * stride + + # Cache images into memory for faster training (WARNING: large datasets may exceed system RAM) + self.imgs = [None] * n + if cache_images: + if cache_images == 'disk': + self.im_cache_dir = Path(Path(self.img_files[0]).parent.as_posix() + '_npy') + self.img_npy = [self.im_cache_dir / Path(f).with_suffix('.npy').name for f in self.img_files] + self.im_cache_dir.mkdir(parents=True, exist_ok=True) + gb = 0 # Gigabytes of cached images + self.img_hw0, self.img_hw = [None] * n, [None] * n + results = ThreadPool(8).imap(lambda x: load_image(*x), zip(repeat(self), range(n))) + pbar = tqdm(enumerate(results), total=n) + for i, x in pbar: + if cache_images == 'disk': + if not self.img_npy[i].exists(): + np.save(self.img_npy[i].as_posix(), x[0]) + gb += self.img_npy[i].stat().st_size + else: + self.imgs[i], self.img_hw0[i], self.img_hw[i] = x + gb += self.imgs[i].nbytes + pbar.desc = f'{prefix}Caching images ({gb / 1E9:.1f}GB)' + pbar.close() + + def cache_labels(self, path=Path('./labels.cache'), prefix=''): + # Cache dataset labels, check images and read shapes + x = {} # dict + nm, nf, ne, nc = 0, 0, 0, 0 # number missing, found, empty, duplicate + pbar = tqdm(zip(self.img_files, self.label_files), desc='Scanning images', total=len(self.img_files)) + for i, (im_file, lb_file) in enumerate(pbar): + try: + # verify images + im = Image.open(im_file) + im.verify() # PIL verify + shape = exif_size(im) # image size + segments = [] # instance segments + assert (shape[0] > 9) & (shape[1] > 9), f'image size {shape} <10 pixels' + assert im.format.lower() in img_formats, f'invalid image format {im.format}' + + # verify labels + if os.path.isfile(lb_file): + nf += 1 # label found + with open(lb_file, 'r') as f: + l = [x.split() for x in f.read().strip().splitlines()] + if any([len(x) > 8 for x in l]): # is segment + classes = np.array([x[0] for x in l], dtype=np.float32) + segments = [np.array(x[1:], dtype=np.float32).reshape(-1, 2) for x in l] # (cls, xy1...) + l = np.concatenate((classes.reshape(-1, 1), segments2boxes(segments)), 1) # (cls, xywh) + l = np.array(l, dtype=np.float32) + if len(l): + assert l.shape[1] == 5, 'labels require 5 columns each' + assert (l >= 0).all(), 'negative labels' + assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels' + assert np.unique(l, axis=0).shape[0] == l.shape[0], 'duplicate labels' + else: + ne += 1 # label empty + l = np.zeros((0, 5), dtype=np.float32) + else: + nm += 1 # label missing + l = np.zeros((0, 5), dtype=np.float32) + x[im_file] = [l, shape, segments] + except Exception as e: + nc += 1 + print(f'{prefix}WARNING: Ignoring corrupted image and/or label {im_file}: {e}') + + pbar.desc = f"{prefix}Scanning '{path.parent / path.stem}' images and labels... " \ + f"{nf} found, {nm} missing, {ne} empty, {nc} corrupted" + pbar.close() + + if nf == 0: + print(f'{prefix}WARNING: No labels found in {path}. See {help_url}') + + x['hash'] = get_hash(self.label_files + self.img_files) + x['results'] = nf, nm, ne, nc, i + 1 + x['version'] = 0.1 # cache version + torch.save(x, path) # save for next time + logging.info(f'{prefix}New cache created: {path}') + return x + + def __len__(self): + return len(self.img_files) + + # def __iter__(self): + # self.count = -1 + # print('ran dataset iter') + # #self.shuffled_vector = np.random.permutation(self.nF) if self.augment else np.arange(self.nF) + # return self + + def __getitem__(self, index): + index = self.indices[index] # linear, shuffled, or image_weights + + hyp = self.hyp + mosaic = self.mosaic and random.random() < hyp['mosaic'] + if mosaic: + # Load mosaic + if random.random() < 0.8: + img, labels = load_mosaic(self, index) + else: + img, labels = load_mosaic9(self, index) + shapes = None + + # MixUp https://arxiv.org/pdf/1710.09412.pdf + if random.random() < hyp['mixup']: + if random.random() < 0.8: + img2, labels2 = load_mosaic(self, random.randint(0, len(self.labels) - 1)) + else: + img2, labels2 = load_mosaic9(self, random.randint(0, len(self.labels) - 1)) + r = np.random.beta(8.0, 8.0) # mixup ratio, alpha=beta=8.0 + img = (img * r + img2 * (1 - r)).astype(np.uint8) + labels = np.concatenate((labels, labels2), 0) + + else: + # Load image + img, (h0, w0), (h, w) = load_image(self, index) + + # Letterbox + shape = self.batch_shapes[self.batch[index]] if self.rect else self.img_size # final letterboxed shape + img, ratio, pad = letterbox(img, shape, auto=False, scaleup=self.augment) + shapes = (h0, w0), ((h / h0, w / w0), pad) # for COCO mAP rescaling + + labels = self.labels[index].copy() + if labels.size: # normalized xywh to pixel xyxy format + labels[:, 1:] = xywhn2xyxy(labels[:, 1:], ratio[0] * w, ratio[1] * h, padw=pad[0], padh=pad[1]) + + if self.augment: + # Augment imagespace + if not mosaic: + img, labels = random_perspective(img, labels, + degrees=hyp['degrees'], + translate=hyp['translate'], + scale=hyp['scale'], + shear=hyp['shear'], + perspective=hyp['perspective']) + + + #img, labels = self.albumentations(img, labels) + + # Augment colorspace + augment_hsv(img, hgain=hyp['hsv_h'], sgain=hyp['hsv_s'], vgain=hyp['hsv_v']) + + # Apply cutouts + # if random.random() < 0.9: + # labels = cutout(img, labels) + + if random.random() < hyp['paste_in']: + sample_labels, sample_images, sample_masks = [], [], [] + while len(sample_labels) < 30: + sample_labels_, sample_images_, sample_masks_ = load_samples(self, random.randint(0, len(self.labels) - 1)) + sample_labels += sample_labels_ + sample_images += sample_images_ + sample_masks += sample_masks_ + #print(len(sample_labels)) + if len(sample_labels) == 0: + break + labels = pastein(img, labels, sample_labels, sample_images, sample_masks) + + nL = len(labels) # number of labels + if nL: + labels[:, 1:5] = xyxy2xywh(labels[:, 1:5]) # convert xyxy to xywh + labels[:, [2, 4]] /= img.shape[0] # normalized height 0-1 + labels[:, [1, 3]] /= img.shape[1] # normalized width 0-1 + + if self.augment: + # flip up-down + if random.random() < hyp['flipud']: + img = np.flipud(img) + if nL: + labels[:, 2] = 1 - labels[:, 2] + + # flip left-right + if random.random() < hyp['fliplr']: + img = np.fliplr(img) + if nL: + labels[:, 1] = 1 - labels[:, 1] + + labels_out = torch.zeros((nL, 6)) + if nL: + labels_out[:, 1:] = torch.from_numpy(labels) + + # Convert + img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416 + img = np.ascontiguousarray(img) + + return torch.from_numpy(img), labels_out, self.img_files[index], shapes + + @staticmethod + def collate_fn(batch): + img, label, path, shapes = zip(*batch) # transposed + for i, l in enumerate(label): + l[:, 0] = i # add target image index for build_targets() + return torch.stack(img, 0), torch.cat(label, 0), path, shapes + + @staticmethod + def collate_fn4(batch): + img, label, path, shapes = zip(*batch) # transposed + n = len(shapes) // 4 + img4, label4, path4, shapes4 = [], [], path[:n], shapes[:n] + + ho = torch.tensor([[0., 0, 0, 1, 0, 0]]) + wo = torch.tensor([[0., 0, 1, 0, 0, 0]]) + s = torch.tensor([[1, 1, .5, .5, .5, .5]]) # scale + for i in range(n): # zidane torch.zeros(16,3,720,1280) # BCHW + i *= 4 + if random.random() < 0.5: + im = F.interpolate(img[i].unsqueeze(0).float(), scale_factor=2., mode='bilinear', align_corners=False)[ + 0].type(img[i].type()) + l = label[i] + else: + im = torch.cat((torch.cat((img[i], img[i + 1]), 1), torch.cat((img[i + 2], img[i + 3]), 1)), 2) + l = torch.cat((label[i], label[i + 1] + ho, label[i + 2] + wo, label[i + 3] + ho + wo), 0) * s + img4.append(im) + label4.append(l) + + for i, l in enumerate(label4): + l[:, 0] = i # add target image index for build_targets() + + return torch.stack(img4, 0), torch.cat(label4, 0), path4, shapes4 + + +# Ancillary functions -------------------------------------------------------------------------------------------------- +def load_image(self, index): + # loads 1 image from dataset, returns img, original hw, resized hw + img = self.imgs[index] + if img is None: # not cached + path = self.img_files[index] + img = cv2.imread(path) # BGR + assert img is not None, 'Image Not Found ' + path + h0, w0 = img.shape[:2] # orig hw + r = self.img_size / max(h0, w0) # resize image to img_size + if r != 1: # always resize down, only resize up if training with augmentation + interp = cv2.INTER_AREA if r < 1 and not self.augment else cv2.INTER_LINEAR + img = cv2.resize(img, (int(w0 * r), int(h0 * r)), interpolation=interp) + return img, (h0, w0), img.shape[:2] # img, hw_original, hw_resized + else: + return self.imgs[index], self.img_hw0[index], self.img_hw[index] # img, hw_original, hw_resized + + +def augment_hsv(img, hgain=0.5, sgain=0.5, vgain=0.5): + r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1 # random gains + hue, sat, val = cv2.split(cv2.cvtColor(img, cv2.COLOR_BGR2HSV)) + dtype = img.dtype # uint8 + + x = np.arange(0, 256, dtype=np.int16) + lut_hue = ((x * r[0]) % 180).astype(dtype) + lut_sat = np.clip(x * r[1], 0, 255).astype(dtype) + lut_val = np.clip(x * r[2], 0, 255).astype(dtype) + + img_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val))).astype(dtype) + cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img) # no return needed + + +def hist_equalize(img, clahe=True, bgr=False): + # Equalize histogram on BGR image 'img' with img.shape(n,m,3) and range 0-255 + yuv = cv2.cvtColor(img, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV) + if clahe: + c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8)) + yuv[:, :, 0] = c.apply(yuv[:, :, 0]) + else: + yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0]) # equalize Y channel histogram + return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB) # convert YUV image to RGB + + +def load_mosaic(self, index): + # loads images in a 4-mosaic + + labels4, segments4 = [], [] + s = self.img_size + yc, xc = [int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border] # mosaic center x, y + indices = [index] + random.choices(self.indices, k=3) # 3 additional image indices + for i, index in enumerate(indices): + # Load image + img, _, (h, w) = load_image(self, index) + + # place img in img4 + if i == 0: # top left + img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles + x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc # xmin, ymin, xmax, ymax (large image) + x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h # xmin, ymin, xmax, ymax (small image) + elif i == 1: # top right + x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc + x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h + elif i == 2: # bottom left + x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h) + x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h) + elif i == 3: # bottom right + x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h) + x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h) + + img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax] + padw = x1a - x1b + padh = y1a - y1b + + # Labels + labels, segments = self.labels[index].copy(), self.segments[index].copy() + if labels.size: + labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padw, padh) # normalized xywh to pixel xyxy format + segments = [xyn2xy(x, w, h, padw, padh) for x in segments] + labels4.append(labels) + segments4.extend(segments) + + # Concat/clip labels + labels4 = np.concatenate(labels4, 0) + for x in (labels4[:, 1:], *segments4): + np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective() + # img4, labels4 = replicate(img4, labels4) # replicate + + # Augment + #img4, labels4, segments4 = remove_background(img4, labels4, segments4) + #sample_segments(img4, labels4, segments4, probability=self.hyp['copy_paste']) + img4, labels4, segments4 = copy_paste(img4, labels4, segments4, probability=self.hyp['copy_paste']) + img4, labels4 = random_perspective(img4, labels4, segments4, + degrees=self.hyp['degrees'], + translate=self.hyp['translate'], + scale=self.hyp['scale'], + shear=self.hyp['shear'], + perspective=self.hyp['perspective'], + border=self.mosaic_border) # border to remove + + return img4, labels4 + + +def load_mosaic9(self, index): + # loads images in a 9-mosaic + + labels9, segments9 = [], [] + s = self.img_size + indices = [index] + random.choices(self.indices, k=8) # 8 additional image indices + for i, index in enumerate(indices): + # Load image + img, _, (h, w) = load_image(self, index) + + # place img in img9 + if i == 0: # center + img9 = np.full((s * 3, s * 3, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles + h0, w0 = h, w + c = s, s, s + w, s + h # xmin, ymin, xmax, ymax (base) coordinates + elif i == 1: # top + c = s, s - h, s + w, s + elif i == 2: # top right + c = s + wp, s - h, s + wp + w, s + elif i == 3: # right + c = s + w0, s, s + w0 + w, s + h + elif i == 4: # bottom right + c = s + w0, s + hp, s + w0 + w, s + hp + h + elif i == 5: # bottom + c = s + w0 - w, s + h0, s + w0, s + h0 + h + elif i == 6: # bottom left + c = s + w0 - wp - w, s + h0, s + w0 - wp, s + h0 + h + elif i == 7: # left + c = s - w, s + h0 - h, s, s + h0 + elif i == 8: # top left + c = s - w, s + h0 - hp - h, s, s + h0 - hp + + padx, pady = c[:2] + x1, y1, x2, y2 = [max(x, 0) for x in c] # allocate coords + + # Labels + labels, segments = self.labels[index].copy(), self.segments[index].copy() + if labels.size: + labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padx, pady) # normalized xywh to pixel xyxy format + segments = [xyn2xy(x, w, h, padx, pady) for x in segments] + labels9.append(labels) + segments9.extend(segments) + + # Image + img9[y1:y2, x1:x2] = img[y1 - pady:, x1 - padx:] # img9[ymin:ymax, xmin:xmax] + hp, wp = h, w # height, width previous + + # Offset + yc, xc = [int(random.uniform(0, s)) for _ in self.mosaic_border] # mosaic center x, y + img9 = img9[yc:yc + 2 * s, xc:xc + 2 * s] + + # Concat/clip labels + labels9 = np.concatenate(labels9, 0) + labels9[:, [1, 3]] -= xc + labels9[:, [2, 4]] -= yc + c = np.array([xc, yc]) # centers + segments9 = [x - c for x in segments9] + + for x in (labels9[:, 1:], *segments9): + np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective() + # img9, labels9 = replicate(img9, labels9) # replicate + + # Augment + #img9, labels9, segments9 = remove_background(img9, labels9, segments9) + img9, labels9, segments9 = copy_paste(img9, labels9, segments9, probability=self.hyp['copy_paste']) + img9, labels9 = random_perspective(img9, labels9, segments9, + degrees=self.hyp['degrees'], + translate=self.hyp['translate'], + scale=self.hyp['scale'], + shear=self.hyp['shear'], + perspective=self.hyp['perspective'], + border=self.mosaic_border) # border to remove + + return img9, labels9 + + +def load_samples(self, index): + # loads images in a 4-mosaic + + labels4, segments4 = [], [] + s = self.img_size + yc, xc = [int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border] # mosaic center x, y + indices = [index] + random.choices(self.indices, k=3) # 3 additional image indices + for i, index in enumerate(indices): + # Load image + img, _, (h, w) = load_image(self, index) + + # place img in img4 + if i == 0: # top left + img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles + x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc # xmin, ymin, xmax, ymax (large image) + x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h # xmin, ymin, xmax, ymax (small image) + elif i == 1: # top right + x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc + x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h + elif i == 2: # bottom left + x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h) + x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h) + elif i == 3: # bottom right + x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h) + x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h) + + img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax] + padw = x1a - x1b + padh = y1a - y1b + + # Labels + labels, segments = self.labels[index].copy(), self.segments[index].copy() + if labels.size: + labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padw, padh) # normalized xywh to pixel xyxy format + segments = [xyn2xy(x, w, h, padw, padh) for x in segments] + labels4.append(labels) + segments4.extend(segments) + + # Concat/clip labels + labels4 = np.concatenate(labels4, 0) + for x in (labels4[:, 1:], *segments4): + np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective() + # img4, labels4 = replicate(img4, labels4) # replicate + + # Augment + #img4, labels4, segments4 = remove_background(img4, labels4, segments4) + sample_labels, sample_images, sample_masks = sample_segments(img4, labels4, segments4, probability=0.5) + + return sample_labels, sample_images, sample_masks + + +def copy_paste(img, labels, segments, probability=0.5): + # Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy) + n = len(segments) + if probability and n: + h, w, c = img.shape # height, width, channels + im_new = np.zeros(img.shape, np.uint8) + for j in random.sample(range(n), k=round(probability * n)): + l, s = labels[j], segments[j] + box = w - l[3], l[2], w - l[1], l[4] + ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area + if (ioa < 0.30).all(): # allow 30% obscuration of existing labels + labels = np.concatenate((labels, [[l[0], *box]]), 0) + segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1)) + cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED) + + result = cv2.bitwise_and(src1=img, src2=im_new) + result = cv2.flip(result, 1) # augment segments (flip left-right) + i = result > 0 # pixels to replace + # i[:, :] = result.max(2).reshape(h, w, 1) # act over ch + img[i] = result[i] # cv2.imwrite('debug.jpg', img) # debug + + return img, labels, segments + + +def remove_background(img, labels, segments): + # Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy) + n = len(segments) + h, w, c = img.shape # height, width, channels + im_new = np.zeros(img.shape, np.uint8) + img_new = np.ones(img.shape, np.uint8) * 114 + for j in range(n): + cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED) + + result = cv2.bitwise_and(src1=img, src2=im_new) + + i = result > 0 # pixels to replace + img_new[i] = result[i] # cv2.imwrite('debug.jpg', img) # debug + + return img_new, labels, segments + + +def sample_segments(img, labels, segments, probability=0.5): + # Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy) + n = len(segments) + sample_labels = [] + sample_images = [] + sample_masks = [] + if probability and n: + h, w, c = img.shape # height, width, channels + for j in random.sample(range(n), k=round(probability * n)): + l, s = labels[j], segments[j] + box = l[1].astype(int).clip(0,w-1), l[2].astype(int).clip(0,h-1), l[3].astype(int).clip(0,w-1), l[4].astype(int).clip(0,h-1) + + #print(box) + if (box[2] <= box[0]) or (box[3] <= box[1]): + continue + + sample_labels.append(l[0]) + + mask = np.zeros(img.shape, np.uint8) + + cv2.drawContours(mask, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED) + sample_masks.append(mask[box[1]:box[3],box[0]:box[2],:]) + + result = cv2.bitwise_and(src1=img, src2=mask) + i = result > 0 # pixels to replace + mask[i] = result[i] # cv2.imwrite('debug.jpg', img) # debug + #print(box) + sample_images.append(mask[box[1]:box[3],box[0]:box[2],:]) + + return sample_labels, sample_images, sample_masks + + +def replicate(img, labels): + # Replicate labels + h, w = img.shape[:2] + boxes = labels[:, 1:].astype(int) + x1, y1, x2, y2 = boxes.T + s = ((x2 - x1) + (y2 - y1)) / 2 # side length (pixels) + for i in s.argsort()[:round(s.size * 0.5)]: # smallest indices + x1b, y1b, x2b, y2b = boxes[i] + bh, bw = y2b - y1b, x2b - x1b + yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw)) # offset x, y + x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh] + img[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax] + labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0) + + return img, labels + + +def letterbox(img, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32): + # Resize and pad image while meeting stride-multiple constraints + shape = img.shape[:2] # current shape [height, width] + if isinstance(new_shape, int): + new_shape = (new_shape, new_shape) + + # Scale ratio (new / old) + r = min(new_shape[0] / shape[0], new_shape[1] / shape[1]) + if not scaleup: # only scale down, do not scale up (for better test mAP) + r = min(r, 1.0) + + # Compute padding + ratio = r, r # width, height ratios + new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r)) + dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1] # wh padding + if auto: # minimum rectangle + dw, dh = np.mod(dw, stride), np.mod(dh, stride) # wh padding + elif scaleFill: # stretch + dw, dh = 0.0, 0.0 + new_unpad = (new_shape[1], new_shape[0]) + ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # width, height ratios + + dw /= 2 # divide padding into 2 sides + dh /= 2 + + if shape[::-1] != new_unpad: # resize + img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR) + top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1)) + left, right = int(round(dw - 0.1)), int(round(dw + 0.1)) + img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color) # add border + return img, ratio, (dw, dh) + + +def random_perspective(img, targets=(), segments=(), degrees=10, translate=.1, scale=.1, shear=10, perspective=0.0, + border=(0, 0)): + # torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(.1, .1), scale=(.9, 1.1), shear=(-10, 10)) + # targets = [cls, xyxy] + + height = img.shape[0] + border[0] * 2 # shape(h,w,c) + width = img.shape[1] + border[1] * 2 + + # Center + C = np.eye(3) + C[0, 2] = -img.shape[1] / 2 # x translation (pixels) + C[1, 2] = -img.shape[0] / 2 # y translation (pixels) + + # Perspective + P = np.eye(3) + P[2, 0] = random.uniform(-perspective, perspective) # x perspective (about y) + P[2, 1] = random.uniform(-perspective, perspective) # y perspective (about x) + + # Rotation and Scale + R = np.eye(3) + a = random.uniform(-degrees, degrees) + # a += random.choice([-180, -90, 0, 90]) # add 90deg rotations to small rotations + s = random.uniform(1 - scale, 1.1 + scale) + # s = 2 ** random.uniform(-scale, scale) + R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s) + + # Shear + S = np.eye(3) + S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # x shear (deg) + S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # y shear (deg) + + # Translation + T = np.eye(3) + T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width # x translation (pixels) + T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height # y translation (pixels) + + # Combined rotation matrix + M = T @ S @ R @ P @ C # order of operations (right to left) is IMPORTANT + if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any(): # image changed + if perspective: + img = cv2.warpPerspective(img, M, dsize=(width, height), borderValue=(114, 114, 114)) + else: # affine + img = cv2.warpAffine(img, M[:2], dsize=(width, height), borderValue=(114, 114, 114)) + + # Visualize + # import matplotlib.pyplot as plt + # ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel() + # ax[0].imshow(img[:, :, ::-1]) # base + # ax[1].imshow(img2[:, :, ::-1]) # warped + + # Transform label coordinates + n = len(targets) + if n: + use_segments = any(x.any() for x in segments) + new = np.zeros((n, 4)) + if use_segments: # warp segments + segments = resample_segments(segments) # upsample + for i, segment in enumerate(segments): + xy = np.ones((len(segment), 3)) + xy[:, :2] = segment + xy = xy @ M.T # transform + xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2] # perspective rescale or affine + + # clip + new[i] = segment2box(xy, width, height) + + else: # warp boxes + xy = np.ones((n * 4, 3)) + xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1 + xy = xy @ M.T # transform + xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8) # perspective rescale or affine + + # create new boxes + x = xy[:, [0, 2, 4, 6]] + y = xy[:, [1, 3, 5, 7]] + new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T + + # clip + new[:, [0, 2]] = new[:, [0, 2]].clip(0, width) + new[:, [1, 3]] = new[:, [1, 3]].clip(0, height) + + # filter candidates + i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10) + targets = targets[i] + targets[:, 1:5] = new[i] + + return img, targets + + +def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n) + # Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio + w1, h1 = box1[2] - box1[0], box1[3] - box1[1] + w2, h2 = box2[2] - box2[0], box2[3] - box2[1] + ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps)) # aspect ratio + return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr) # candidates + + +def bbox_ioa(box1, box2): + # Returns the intersection over box2 area given box1, box2. box1 is 4, box2 is nx4. boxes are x1y1x2y2 + box2 = box2.transpose() + + # Get the coordinates of bounding boxes + b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3] + b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3] + + # Intersection area + inter_area = (np.minimum(b1_x2, b2_x2) - np.maximum(b1_x1, b2_x1)).clip(0) * \ + (np.minimum(b1_y2, b2_y2) - np.maximum(b1_y1, b2_y1)).clip(0) + + # box2 area + box2_area = (b2_x2 - b2_x1) * (b2_y2 - b2_y1) + 1e-16 + + # Intersection over box2 area + return inter_area / box2_area + + +def cutout(image, labels): + # Applies image cutout augmentation https://arxiv.org/abs/1708.04552 + h, w = image.shape[:2] + + # create random masks + scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16 # image size fraction + for s in scales: + mask_h = random.randint(1, int(h * s)) + mask_w = random.randint(1, int(w * s)) + + # box + xmin = max(0, random.randint(0, w) - mask_w // 2) + ymin = max(0, random.randint(0, h) - mask_h // 2) + xmax = min(w, xmin + mask_w) + ymax = min(h, ymin + mask_h) + + # apply random color mask + image[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)] + + # return unobscured labels + if len(labels) and s > 0.03: + box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32) + ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area + labels = labels[ioa < 0.60] # remove >60% obscured labels + + return labels + + +def pastein(image, labels, sample_labels, sample_images, sample_masks): + # Applies image cutout augmentation https://arxiv.org/abs/1708.04552 + h, w = image.shape[:2] + + # create random masks + scales = [0.75] * 2 + [0.5] * 4 + [0.25] * 4 + [0.125] * 4 + [0.0625] * 6 # image size fraction + for s in scales: + if random.random() < 0.2: + continue + mask_h = random.randint(1, int(h * s)) + mask_w = random.randint(1, int(w * s)) + + # box + xmin = max(0, random.randint(0, w) - mask_w // 2) + ymin = max(0, random.randint(0, h) - mask_h // 2) + xmax = min(w, xmin + mask_w) + ymax = min(h, ymin + mask_h) + + box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32) + if len(labels): + ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area + else: + ioa = np.zeros(1) + + if (ioa < 0.30).all() and len(sample_labels) and (xmax > xmin+20) and (ymax > ymin+20): # allow 30% obscuration of existing labels + sel_ind = random.randint(0, len(sample_labels)-1) + #print(len(sample_labels)) + #print(sel_ind) + #print((xmax-xmin, ymax-ymin)) + #print(image[ymin:ymax, xmin:xmax].shape) + #print([[sample_labels[sel_ind], *box]]) + #print(labels.shape) + hs, ws, cs = sample_images[sel_ind].shape + r_scale = min((ymax-ymin)/hs, (xmax-xmin)/ws) + r_w = int(ws*r_scale) + r_h = int(hs*r_scale) + + if (r_w > 10) and (r_h > 10): + r_mask = cv2.resize(sample_masks[sel_ind], (r_w, r_h)) + r_image = cv2.resize(sample_images[sel_ind], (r_w, r_h)) + temp_crop = image[ymin:ymin+r_h, xmin:xmin+r_w] + m_ind = r_mask > 0 + if m_ind.astype(np.int).sum() > 60: + temp_crop[m_ind] = r_image[m_ind] + #print(sample_labels[sel_ind]) + #print(sample_images[sel_ind].shape) + #print(temp_crop.shape) + box = np.array([xmin, ymin, xmin+r_w, ymin+r_h], dtype=np.float32) + if len(labels): + labels = np.concatenate((labels, [[sample_labels[sel_ind], *box]]), 0) + else: + labels = np.array([[sample_labels[sel_ind], *box]]) + + image[ymin:ymin+r_h, xmin:xmin+r_w] = temp_crop + + return labels + +class Albumentations: + # YOLOv5 Albumentations class (optional, only used if package is installed) + def __init__(self): + self.transform = None + import albumentations as A + + self.transform = A.Compose([ + A.CLAHE(p=0.01), + A.RandomBrightnessContrast(brightness_limit=0.2, contrast_limit=0.2, p=0.01), + A.RandomGamma(gamma_limit=[80, 120], p=0.01), + A.Blur(p=0.01), + A.MedianBlur(p=0.01), + A.ToGray(p=0.01), + A.ImageCompression(quality_lower=75, p=0.01),], + bbox_params=A.BboxParams(format='pascal_voc', label_fields=['class_labels'])) + + #logging.info(colorstr('albumentations: ') + ', '.join(f'{x}' for x in self.transform.transforms if x.p)) + + def __call__(self, im, labels, p=1.0): + if self.transform and random.random() < p: + new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0]) # transformed + im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])]) + return im, labels + + +def create_folder(path='./new'): + # Create folder + if os.path.exists(path): + shutil.rmtree(path) # delete output folder + os.makedirs(path) # make new output folder + + +def flatten_recursive(path='../coco'): + # Flatten a recursive directory by bringing all files to top level + new_path = Path(path + '_flat') + create_folder(new_path) + for file in tqdm(glob.glob(str(Path(path)) + '/**/*.*', recursive=True)): + shutil.copyfile(file, new_path / Path(file).name) + + +def extract_boxes(path='../coco/'): # from utils.datasets import *; extract_boxes('../coco128') + # Convert detection dataset into classification dataset, with one directory per class + + path = Path(path) # images dir + shutil.rmtree(path / 'classifier') if (path / 'classifier').is_dir() else None # remove existing + files = list(path.rglob('*.*')) + n = len(files) # number of files + for im_file in tqdm(files, total=n): + if im_file.suffix[1:] in img_formats: + # image + im = cv2.imread(str(im_file))[..., ::-1] # BGR to RGB + h, w = im.shape[:2] + + # labels + lb_file = Path(img2label_paths([str(im_file)])[0]) + if Path(lb_file).exists(): + with open(lb_file, 'r') as f: + lb = np.array([x.split() for x in f.read().strip().splitlines()], dtype=np.float32) # labels + + for j, x in enumerate(lb): + c = int(x[0]) # class + f = (path / 'classifier') / f'{c}' / f'{path.stem}_{im_file.stem}_{j}.jpg' # new filename + if not f.parent.is_dir(): + f.parent.mkdir(parents=True) + + b = x[1:] * [w, h, w, h] # box + # b[2:] = b[2:].max() # rectangle to square + b[2:] = b[2:] * 1.2 + 3 # pad + b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int) + + b[[0, 2]] = np.clip(b[[0, 2]], 0, w) # clip boxes outside of image + b[[1, 3]] = np.clip(b[[1, 3]], 0, h) + assert cv2.imwrite(str(f), im[b[1]:b[3], b[0]:b[2]]), f'box failure in {f}' + + +def autosplit(path='../coco', weights=(0.9, 0.1, 0.0), annotated_only=False): + """ Autosplit a dataset into train/val/test splits and save path/autosplit_*.txt files + Usage: from utils.datasets import *; autosplit('../coco') + Arguments + path: Path to images directory + weights: Train, val, test weights (list) + annotated_only: Only use images with an annotated txt file + """ + path = Path(path) # images dir + files = sum([list(path.rglob(f"*.{img_ext}")) for img_ext in img_formats], []) # image files only + n = len(files) # number of files + indices = random.choices([0, 1, 2], weights=weights, k=n) # assign each image to a split + + txt = ['autosplit_train.txt', 'autosplit_val.txt', 'autosplit_test.txt'] # 3 txt files + [(path / x).unlink() for x in txt if (path / x).exists()] # remove existing + + print(f'Autosplitting images from {path}' + ', using *.txt labeled images only' * annotated_only) + for i, img in tqdm(zip(indices, files), total=n): + if not annotated_only or Path(img2label_paths([str(img)])[0]).exists(): # check label + with open(path / txt[i], 'a') as f: + f.write(str(img) + '\n') # add image to txt file + + +def load_segmentations(self, index): + key = '/work/handsomejw66/coco17/' + self.img_files[index] + #print(key) + # /work/handsomejw66/coco17/ + return self.segs[key] diff --git a/ai/detector_rev2.yaml b/ai/detector_rev2.yaml new file mode 100644 index 0000000..6537433 --- /dev/null +++ b/ai/detector_rev2.yaml @@ -0,0 +1,10 @@ +train: /data/dataset/detector_train.txt +val: /data/dataset/detector_val.txt +test: /data/dataset/detector_test.txt + +nc: 12 + +names: ['Airbus A319', 'Airbus A320', 'Airbus A321', + 'Beechcraft Bonanza', 'Boeing 737', 'Bombardier CL-600', + 'Cessna Citation', 'Cessna Skyhawk', 'Cessna Skylane', + 'Embraer ERJ-170', 'Piper PA-28 Cherokee', 'Piper PA-32 Cherokee Six'] diff --git a/ai/localizer_rev2.yaml b/ai/localizer_rev2.yaml new file mode 100644 index 0000000..31396a2 --- /dev/null +++ b/ai/localizer_rev2.yaml @@ -0,0 +1,9 @@ +train: /data/dataset/localizer_train +val: /data/dataset/localizer_val +test: [/data/dataset/detector_train, /data/dataset/detector_val] +test2: /data/dataset/localizer_test +test3: /data/dataset/detector_test + +nc: 1 + +names: ['aircraft'] diff --git a/ai/metrics2.py b/ai/metrics2.py new file mode 100644 index 0000000..61cd550 --- /dev/null +++ b/ai/metrics2.py @@ -0,0 +1,280 @@ +# Model validation metrics + +from pathlib import Path + +import matplotlib.pyplot as plt +import numpy as np +import torch +import sys +import math + +from . import general + + +def fitness(x): + # Model fitness as a weighted combination of metrics + w = [0.0, 0.0, 0.1, 0.9] # weights for [P, R, mAP@0.5, mAP@0.5:0.95] + return (x[:, :4] * w).sum(1) + + +def ap_per_class(tp, conf, pred_cls, target_cls, v5_metric=False, plot=False, save_dir='.', names=()): + """ Compute the average precision, given the recall and precision curves. + Source: https://github.com/rafaelpadilla/Object-Detection-Metrics. + # Arguments + tp: True positives (nparray, nx1 or nx10). + conf: Objectness value from 0-1 (nparray). + pred_cls: Predicted object classes (nparray). + target_cls: True object classes (nparray). + plot: Plot precision-recall curve at mAP@0.5 + save_dir: Plot save directory + # Returns + The average precision as computed in py-faster-rcnn. + """ + + # Sort by objectness + i = np.argsort(-conf) + tp, conf, pred_cls = tp[i], conf[i], pred_cls[i] + + # Find unique classes + unique_classes = np.unique(target_cls) + nc = unique_classes.shape[0] # number of classes, number of detections + + # Create Precision-Recall curve and compute AP for each class + px, py = np.linspace(0, 1, 1000), [] # for plotting + ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000)) + for ci, c in enumerate(unique_classes): + i = pred_cls == c + n_l = (target_cls == c).sum() # number of labels + n_p = i.sum() # number of predictions + + if n_p == 0 or n_l == 0: + continue + else: + # Accumulate FPs and TPs + fpc = (1 - tp[i]).cumsum(0) + tpc = tp[i].cumsum(0) + + # Recall + recall = tpc / (n_l + 1e-16) # recall curve + r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases + + # Precision + precision = tpc / (tpc + fpc) # precision curve + p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score + + # AP from recall-precision curve + for j in range(tp.shape[1]): + ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j], v5_metric=v5_metric) + if plot and j == 0: + py.append(np.interp(px, mrec, mpre)) # precision at mAP@0.5 + + # Compute F1 (harmonic mean of precision and recall) + f1 = 2 * p * r / (p + r + 1e-16) + if plot: + plot_pr_curve(px, py, ap, Path(save_dir) / 'PR_curve.png', names) + plot_mc_curve(px, f1, Path(save_dir) / 'F1_curve.png', names, ylabel='F1') + plot_mc_curve(px, p, Path(save_dir) / 'P_curve.png', names, ylabel='Precision') + plot_mc_curve(px, r, Path(save_dir) / 'R_curve.png', names, ylabel='Recall') + + i = f1.mean(0).argmax() # max F1 index + return p[:, i], r[:, i], ap, f1[:, i], unique_classes.astype('int32') + + +def compute_ap(recall, precision, v5_metric=False): + """ Compute the average precision, given the recall and precision curves + # Arguments + recall: The recall curve (list) + precision: The precision curve (list) + v5_metric: Assume maximum recall to be 1.0, as in YOLOv5, MMDetetion etc. + # Returns + Average precision, precision curve, recall curve + """ + + # Append sentinel values to beginning and end + if v5_metric: # New YOLOv5 metric, same as MMDetection and Detectron2 repositories + mrec = np.concatenate(([0.], recall, [1.0])) + else: # Old YOLOv5 metric, i.e. default YOLOv7 metric + mrec = np.concatenate(([0.], recall, [recall[-1] + 0.01])) + mpre = np.concatenate(([1.], precision, [0.])) + + # Compute the precision envelope + mpre = np.flip(np.maximum.accumulate(np.flip(mpre))) + + # Integrate area under curve + method = 'interp' # methods: 'continuous', 'interp' + if method == 'interp': + x = np.linspace(0, 1, 101) # 101-point interp (COCO) + ap = np.trapz(np.interp(x, mrec, mpre), x) # integrate + else: # 'continuous' + i = np.where(mrec[1:] != mrec[:-1])[0] # points where x axis (recall) changes + ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) # area under curve + + return ap, mpre, mrec + + +class ConfusionMatrix: + # Updated version of https://github.com/kaanakan/object_detection_confusion_matrix + # Further updates made here + def __init__(self, nc, conf=0.25, iou_thres=0.45): + self.matrix = np.zeros((nc + 1, nc + 1)) + self.nc = nc # number of classes + self.conf = conf + self.iou_thres = iou_thres + + def process_batch(self, detections, labels): + """ + Return intersection-over-union (Jaccard index) of boxes. + Both sets of boxes are expected to be in (x1, y1, x2, y2) format. + Arguments: + detections (Array[N, 6]), x1, y1, x2, y2, conf, class + labels (Array[M, 5]), class, x1, y1, x2, y2 + Returns: + None, updates confusion matrix accordingly + """ + detections = detections[detections[:, 4] > self.conf] + gt_classes = labels[:, 0].int() + detection_classes = detections[:, 5].int() + iou = general.box_iou(labels[:, 1:], detections[:, :4]) + + x = torch.where(iou > self.iou_thres) + if x[0].shape[0]: + matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy() + if x[0].shape[0] > 1: + matches = matches[matches[:, 2].argsort()[::-1]] + matches = matches[np.unique(matches[:, 1], return_index=True)[1]] + matches = matches[matches[:, 2].argsort()[::-1]] + matches = matches[np.unique(matches[:, 0], return_index=True)[1]] + else: + matches = np.zeros((0, 3)) + + n = matches.shape[0] > 0 + m0, m1, _ = matches.transpose().astype(np.int16) + for i, gc in enumerate(gt_classes): + j = m0 == i + if n and sum(j) == 1: + self.matrix[gc, detection_classes[m1[j]]] += 1 # correct + else: + self.matrix[gc, self.nc] += 1 # background FP + + if n: + for i, dc in enumerate(detection_classes): + if not any(m1 == i): + self.matrix[self.nc, dc] += 1 # background FN + + def matrix(self): + return self.matrix + + def plot(self, save_dir='', names=(), normalize=True, transpose=False, + filename=None, round_to_int=False): + try: + import seaborn as sn + + if normalize: + array = self.matrix / (self.matrix.sum(0).reshape(1, self.nc + 1) + 1E-6) # normalize + else: + array = self.matrix.copy() + if transpose: + array = np.transpose(array) + array[array < 0.005] = np.nan # don't annotate (would appear as 0.00) + + fig = plt.figure(figsize=(12, 9), tight_layout=True) + sn.set(font_scale=1.0 if self.nc < 50 else 0.8) # for label size + labels = (0 < len(names) < 99) and len(names) == self.nc # apply names to ticklabels + if not transpose: + xticklabels=names + ['background FN'] if labels else "auto" + yticklabels=names + ['background FP'] if labels else "auto" + else: + xticklabels=names + ['background FP'] if labels else "auto" + yticklabels=names + ['background FN'] if labels else "auto" + if round_to_int: + fmt = '.0f' + else: + fmt = '.2f' + sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, + cmap='Blues', fmt=fmt, square=True, + xticklabels=xticklabels, + yticklabels=yticklabels).set_facecolor((1, 1, 1)) + if not transpose: + fig.axes[0].set_xlabel('Predicted') + fig.axes[0].set_ylabel('True') + else: + fig.axes[0].set_xlabel('True') + fig.axes[0].set_ylabel('Predicted') + if filename is None: + fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250) + else: + fig.savefig(Path(save_dir) / (filename + '.png'), dpi=250) + except Exception as e: + pass + + def print(self): + for i in range(self.nc + 1): + print(' '.join(map(str, self.matrix[i]))) + + def write(self, path): + with open(path, 'w') as wfile: + stdout = sys.stdout + sys.stdout = wfile + self.print() + sys.stdout = stdout + + def read(self, path): + data = np.fromfile(path, sep=' ') + self.nc = round(math.sqrt(len(data))) - 1 + self.matrix = np.reshape(data, (self.nc + 1, self.nc + 1)) + + def norm(self, axis=None): + # Returns a new confusion matrix that's normalized + if axis is None: + m = np.copy(self.matrix) + elif axis == 0: + m = self.matrix / (self.matrix.sum(0).reshape(1, self.nc + 1) + 1E-6) + elif axis == 1: + m = self.matrix / (self.matrix.sum(1).reshape(self.nc + 1, 1) + 1E-6) + else: + raise Exception('! Invalid axis') + cm = ConfusionMatrix(self.nc, self.conf, self.iou_thres) + cm.matrix = m + return cm + + +# Plots ---------------------------------------------------------------------------------------------------------------- + +def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()): + # Precision-recall curve + fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True) + py = np.stack(py, axis=1) + + if 0 < len(names) < 21: # display per-class legend if < 21 classes + for i, y in enumerate(py.T): + ax.plot(px, y, linewidth=1, label=f'{names[i]} {ap[i, 0]:.3f}') # plot(recall, precision) + else: + ax.plot(px, py, linewidth=1, color='grey') # plot(recall, precision) + + ax.plot(px, py.mean(1), linewidth=3, color='blue', label='all classes %.3f mAP@0.5' % ap[:, 0].mean()) + ax.set_xlabel('Recall') + ax.set_ylabel('Precision') + ax.set_xlim(0, 1) + ax.set_ylim(0, 1) + plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left") + fig.savefig(Path(save_dir), dpi=250) + + +def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence', ylabel='Metric'): + # Metric-confidence curve + fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True) + + if 0 < len(names) < 21: # display per-class legend if < 21 classes + for i, y in enumerate(py): + ax.plot(px, y, linewidth=1, label=f'{names[i]}') # plot(confidence, metric) + else: + ax.plot(px, py.T, linewidth=1, color='grey') # plot(confidence, metric) + + y = py.mean(0) + ax.plot(px, y, linewidth=3, color='blue', label=f'all classes {y.max():.2f} at {px[y.argmax()]:.3f}') + ax.set_xlabel(xlabel) + ax.set_ylabel(ylabel) + ax.set_xlim(0, 1) + ax.set_ylim(0, 1) + plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left") + fig.savefig(Path(save_dir), dpi=250) diff --git a/ai/model.py b/ai/model.py new file mode 100755 index 0000000..c37530c --- /dev/null +++ b/ai/model.py @@ -0,0 +1,1116 @@ +#!/usr/bin/env python +# coding: utf-8 + +""" +SkyScan Edge AI +""" + +import os +import sys +import json +import glob +import shutil +import subprocess +import numpy as np +import pandas as pd +import IPython.display + +from utils.metrics2 import ConfusionMatrix + +display_width = 120 +default_seed = 1903 + +class Dataset: + + def create_from_images(self, path, suffix='jpg'): + """ + Initialize dataframe using a directory of images + """ + image_paths = sorted(glob.glob(path + '/**/*.'+suffix, recursive=True)) + self.df = pd.DataFrame(image_paths, columns=['path']) + self.df['filename'] = self.df['path'].apply( + lambda x: os.path.basename(x)) + self.df['icao24'] = self.df['path'].apply( + lambda x: os.path.basename(x).split('_')[0].lower()) + + def create_from_dirs(self, paths, suffix='jpg'): + """ + Initialize dataframe using a list of directories of images + """ + image_paths = [] + for path in paths: + more_image_paths = sorted(glob.glob(path + '/**/*.'+suffix, + recursive=True)) + image_paths.extend(more_image_paths) + self.df = pd.DataFrame(image_paths, columns=['path']) + self.df['filename'] = self.df['path'].apply( + lambda x: os.path.basename(x)) + self.df['icao24'] = self.df['path'].apply( + lambda x: os.path.basename(x).split('_')[0].lower()) + + def create_from_paths_file(self, path, suffix='jpg'): + """ + Initialize dataframe using a file of paths + Note: Assumes paths are relative to the folder from which code is run + """ + #path_file = open(path, 'r') + with open(path, 'r') as path_file: + image_paths = path_file.readlines() + self.df = pd.DataFrame(image_paths, columns=['path']) + self.df['filename'] = self.df['path'].apply( + lambda x: os.path.basename(x)) + self.df['icao24'] = self.df['path'].apply( + lambda x: os.path.basename(x).split('_')[0].lower()) + + def concat(self, second_dataset): + """ + Return a new dataset which is the concatenation of this one + and the provided `second_dataset` + """ + new_dataset = Dataset() + new_dataset.df = pd.concat((self.df, second_dataset.df), + ignore_index=True) + return new_dataset + + def load_makemodel_string(self, database_path=None, method=1): + """ + Add raw make+model string from 3rd-party database + """ + if method == 0: + # Assumes images are already sorted into folders named after + # make+model string + df['string'] = df['path'].apply(lambda x: x.split('/')[-2]) + else: + # Convert icao24 (ICAO 24-bit address) to make+model string with + # OpenSky Network database. Source: + # https://opensky-network.org/datasets/metadata/aircraftDatabase.csv + + # Step 1: Load and format lookup table + database_dataframe = pd.read_csv(database_path, usecols=[ + 'icao24', 'manufacturername', 'model']) + database_dataframe['string'] = database_dataframe[ + 'manufacturername'] + ' ' + database_dataframe['model'] + database_dataframe = database_dataframe[['icao24', 'string']] + + # Step 2: Join tables + self.df = pd.merge(self.df, database_dataframe, + how='left', on='icao24') + database_dataframe = None + + def load_makemodel_mm(self, hierarchy_path, + print_lookup_table=False, warn_missing=True): + """ + Convert raw make+model string to sanitized version with a lookup table + """ + + # Convert class hierarchy to table of make/model/string combinations + # Note: 'string' is unsanitized make+model from database; + # 'mm' is make+model after data cleaning + class_hierarchy = json.load(open(hierarchy_path)) + class_array = np.concatenate([np.concatenate([np.stack([ + np.array([make + ' ' + model, make, model, string]) + for string in class_hierarchy['make_model_strings'][make][model]], axis=0) + for model in class_hierarchy['make_model_strings'][make]], axis=0) + for make in class_hierarchy['make_model_strings']], axis=0) + class_dataframe = pd.DataFrame(class_array, columns=['mm', 'make', 'model', 'string']) + + if print_lookup_table: + with pd.option_context('display.max_rows', None, 'display.max_columns', None, 'display.width', display_width): + print(class_dataframe[['mm', 'string']]) + + # Join class information to dataset item information + self.df = pd.merge(self.df, class_dataframe, how='left', on='string') + + # Check for items not found in class hierarchy + if warn_missing: + missing = sorted(self.df[self.df['mm'].isna() & ~self.df['string'].isna()]['string'].unique()) + if len(missing) > 0: + print('! Make+model strings missing from class hierarchy!') + print(missing) + + # Return class hierarchy + return class_hierarchy + + def print_classes(self, group_field='mm', title=None, condition=None): + """ + Print count of each class + """ + if title is not None: + print('-' * display_width) + print(title) + if condition is not None: + df = self.df[condition] + else: + df = self.df + image_count = df.groupby(group_field)[group_field].count() + plane_count = df.groupby('icao24').head(1).groupby(group_field)[group_field].count() + count_dataframe = pd.concat([image_count, plane_count], axis=1) + count_dataframe.columns=['image_count', 'aircraft_count'] + with pd.option_context('display.max_rows', None, + 'display.width', display_width): + print(count_dataframe) + print() + print('Identified images:', image_count.sum()) + print('Identified aircraft:', plane_count.sum()) + print('Total images:', df['path'].count()) + print('Total aircraft:', df.groupby('icao24')['icao24'].head(1).count()) + + def add_usable(self, detector_classes, block_makes=None): + """ + Mark usable items for localizer and detector + """ + # Identified (and unidentified, if not overwritten below) + if block_makes is not None: + self.df['localizer_usable'] = ~self.df['make'].isin(block_makes) + else: + self.df['localizer_usable'] = True + self.df['detector_usable'] = self.df['mm'].isin(detector_classes) + + # Unidentified (if different from above) + # self.df.loc[self.df['mm'].isna(), 'localizer_usable'] = True + # self.df.loc[self.df['mm'].isna(), 'detector_usable'] = False + + def print(self, title=None, condition=None): + """ + Print dataframe + """ + if title is not None: + print('-' * display_width) + print(title) + with pd.option_context('display.max_columns', None, + 'display.width', display_width): + if condition is None: + print(self.df) + else: + print(self.df[condition]) + + def write_suggested_labels(self, output_path, + max_planes_per_class=3, + max_images_per_plane=5, + seed1=default_seed, + seed2=default_seed): + """ + Write a file with a list of suggested images to hand-label + with bounding boxes. Include N random images per plane from + up to M planes of each class, to emphasize less common classes. + """ + # Step 1: Select planes + planes_to_label = self.df.groupby('icao24').head(1)[['icao24', 'mm']].sample(frac=1, random_state=seed1).groupby('mm').head(max_planes_per_class).sort_values(by='mm') + + # Step 2: Select images + images_to_label = self.df[self.df['icao24'].isin(planes_to_label['icao24'])] + images_to_label = images_to_label[images_to_label['localizer_usable']] + images_to_label = images_to_label.sample(frac=1, random_state=seed2).groupby('icao24').head(max_images_per_plane).sort_values(by=['mm', 'icao24']) + + # Step 3: Save + images_to_label['path'].to_csv(output_path, header=False, index=False) + + def write_suggested_labels_prop(self, output_path, + num_planes=10, + max_images_per_plane=5, + seed1=default_seed, + seed2=default_seed): + """ + Write a file with a list of suggested images to hand-label + with bounding boxes. Include N random images per plane from + each of M planes, to follow the observed distribution of classes. + """ + # Step 1: Select planes + planes_to_label = self.df.groupby('icao24').head(1)[['icao24', 'mm']].sample(frac=1, random_state=seed1).head(num_planes).sort_values(by=['mm', 'icao24']) + + # Step 2: Select images + images_to_label = self.df[self.df['icao24'].isin(planes_to_label['icao24'])] + images_to_label = images_to_label[images_to_label['localizer_usable']] + images_to_label = images_to_label.sample(frac=1, random_state=seed2).groupby('icao24').head(max_images_per_plane).sort_values(by=['mm', 'icao24']) + + # Step 3: Save + images_to_label['path'].to_csv(output_path, header=False, index=False) + + def load_labelbox_labels(self, input_path, + img_height=1080, img_width=1920, min_area=100): + """ + Load manual bounding boxes from LabelBox export file into dataframe + """ + label_struct = json.load(open(input_path)) + + self.df['manual_bbox_flag'] = False + self.df['manual_bbox_count'] = 0 + self.df['manual_bbox_desc'] = '' + filename_index = pd.Index(self.df['filename']) + for item in label_struct: + # Find this item in dataframe + filename = item['External ID'] + try: + loc = filename_index.get_loc(filename) + except KeyError: + print('! Did not find entry for', filename) + continue + + # Parse hand-labeled bounding box + desc = '' + valid_boxes = 0 + for obj in item['Label']['objects']: + bbox = obj['bbox'] + top = bbox['top'] + left = bbox['left'] + height = bbox['height'] + width = bbox['width'] + if height * width >= min_area: + if valid_boxes > 0: + desc += '\n' + desc += '0 %f %f %f %f' % ( + (left + 0.5 * width) / img_width, + (top + 0.5 * height) / img_height, + width / img_width, + height / img_height + ) + valid_boxes += 1 + self.df.at[loc, 'manual_bbox_flag'] = True + self.df.at[loc, 'manual_bbox_count'] = valid_boxes + self.df.at[loc, 'manual_bbox_desc'] = desc + + def load_yolov7_labels(self, name, + img_height=1080, img_width=1920, + min_area=100, + always_flag=['detector_train', 'detector_test']): + """ + Load YOLOv7-generated bounding boxes + """ + path = os.path.join('runs/test', name, 'best_predictions.json') + label_struct = json.load(open(path)) + + self.df['auto_bbox_flag'] = False + self.df['auto_bbox_count'] = 0 + self.df['auto_bbox_desc'] = '' + filename_index = pd.Index(self.df['filename'].str[:-4]) + for entry in label_struct: + # Find this item in dataframe + filename = entry['image_id'] + loc = filename_index.get_loc(filename) + + # Parse auto-labeled bounding box + left, top, width, height = tuple(entry['bbox']) + if height * width >= min_area: + if self.df.loc[loc, 'auto_bbox_count'] > 0: + self.df.loc[loc, 'auto_bbox_desc'] += '\n' + self.df.loc[loc, 'auto_bbox_desc'] += '0 %f %f %f %f' % ( + (left + 0.5 * width) / img_width, + (top + 0.5 * height) / img_height, + width / img_width, + height / img_height + ) + self.df.loc[loc, 'auto_bbox_count'] += 1 + + # Flag all entries for which bboxes were sought + # (e.g., detector train/test), even those with no bboxes found + for colname in always_flag: + self.df.loc[self.df[colname], 'auto_bbox_flag'] = True + + def split_train_test(self, detector_classes, + localizer_train_count = 2, + detector_test_count = 3, + seed3 = default_seed, + seed4 = default_seed, + verbose = True): + """ + Divides images into four categories: localizer training, localizer + testing, detector training, and detector testing. The division is + such that no one plane's images appear in more than one category. + Not all images are assigned, because localizer categories can only + use a plane's hand-labeled bounding boxes. + """ + + # Assign planes to use: train/test for localizer/detector + planes = self.df[['icao24', 'mm', 'manual_bbox_flag']].groupby('icao24', as_index=False).max() + planes_manual = planes[planes['manual_bbox_flag']] + planes_notman = planes[~planes['manual_bbox_flag']] + planes_localizer = planes_manual + planes_localizer_train = planes_localizer.sample(frac=1, random_state=seed3).groupby('mm', as_index=False).head(localizer_train_count) + planes_localizer_test = planes_localizer[~planes_localizer['icao24'].isin(planes_localizer_train['icao24'])] + planes_detector = planes_notman[planes_notman['mm'].isin(detector_classes)] + planes_detector_test = planes_detector.sample(frac=1, random_state=seed4).groupby('mm', as_index=False).head(detector_test_count) + planes_detector_train = planes_detector[~planes_detector['icao24'].isin(planes_detector_test['icao24'])] + + # Print plane counts + if verbose: + print('planes', len(planes)) + print('planes_manual', len(planes_manual)) + print('planes_notman', len(planes_notman)) + print('planes_localizer_train', len(planes_localizer_train)) + print('planes_localizer_test', len(planes_localizer_test)) + print('planes_detector_train', len(planes_detector_train)) + print('planes_detector_test', len(planes_detector_test)) + + # Assign dataset items to use: train/test for localizer/detector + items_localizer_train = self.df[self.df['icao24'].isin(planes_localizer_train['icao24'])] + items_localizer_train = items_localizer_train[items_localizer_train['manual_bbox_flag']] + items_localizer_test = self.df[self.df['icao24'].isin(planes_localizer_test['icao24'])] + items_localizer_test = items_localizer_test[items_localizer_test['manual_bbox_flag']] + items_detector_train = self.df[self.df['icao24'].isin(planes_detector_train['icao24'])] + items_detector_test = self.df[self.df['icao24'].isin(planes_detector_test['icao24'])] + + # Save to main dataframe + self.df['localizer_train'] = self.df['path'].isin(items_localizer_train['path']) + self.df['localizer_test'] = self.df['path'].isin(items_localizer_test['path']) + self.df['detector_train'] = self.df['path'].isin(items_detector_train['path']) + self.df['detector_test'] = self.df['path'].isin(items_detector_test['path']) + + # Print item counts + if verbose: + print('items', self.df.shape[0]) + print('items_localizer_train', self.df['localizer_train'].sum()) + print('items_localizer_test', self.df['localizer_test'].sum()) + print('items_detector_train', self.df['detector_train'].sum()) + print('items_detector_test', self.df['detector_test'].sum()) + + def split_train_val_shared(self, detector_classes, + val_fraction = 0.3, + seed3 = default_seed, + verbose = True): + """ + Assigns images to four categories: localizer training, localizer + validation, detector training, and detector validation. The localizer + images are a proper subset of the detector images, but no image is used + for both training and validation. + """ + + # Assign planes to use: either train or validation + planes = self.df[['icao24', 'mm', 'manual_bbox_flag']].groupby('icao24', as_index=False).max() + val_count = int(val_fraction * len(planes)) + planes_randomized = planes.sample(frac=1, random_state=seed3) + planes_train = planes_randomized.iloc[val_count:] + planes_val = planes_randomized.iloc[:val_count] + + # Assign dataset items to use: train/validation for localizer/detector + items_train = self.df[self.df['icao24'].isin(planes_train['icao24'])] + items_val = self.df[self.df['icao24'].isin(planes_val['icao24'])] + items_localizer_train = items_train[items_train['manual_bbox_flag']] + items_localizer_val = items_val[items_val['manual_bbox_flag']] + items_detector_train = items_train[items_train['mm'].isin(detector_classes)] + items_detector_val = items_val[items_val['mm'].isin(detector_classes)] + + # Save to main dataframe + self.df['localizer_train'] = self.df['path'].isin( + items_localizer_train['path']) + self.df['localizer_val'] = self.df['path'].isin( + items_localizer_val['path']) + self.df['detector_train'] = self.df['path'].isin( + items_detector_train['path']) + self.df['detector_val'] = self.df['path'].isin( + items_detector_val['path']) + + # Print plane counts + if verbose: + print('planes', len(planes)) + print('planes_train', len(planes_train)) + print('planes_val', len(planes_val)) + + # Print item counts + if verbose: + print('items', self.df.shape[0]) + print('items_train', len(items_train)) + print('items_val', len(items_val)) + print('items_localizer_train', self.df['localizer_train'].sum()) + print('items_localizer_val', self.df['localizer_val'].sum()) + print('items_detector_train', self.df['detector_train'].sum()) + print('items_detector_val', self.df['detector_val'].sum()) + + def check_categories_nonintersecting(self, categories): + """ + Confirm that no plane's images appear in multiple usage categories + """ + for cat1idx in range(len(categories)): + cat1 = set(self.df.loc[self.df[categories[cat1idx]], 'icao24']) + for cat2idx in range(cat1idx + 1, len(categories)): + cat2 = set(self.df.loc[self.df[categories[cat2idx]], 'icao24']) + inter = cat1.intersection(cat2) + if len(inter) > 0: + print('! Warning: Category Overlap!', + categories[cat1idx], categories[cat2idx]) + + def write_folders(self, base_path, categories): + """ + Create folders to hold dataset files + """ + for category in categories: + os.makedirs(os.path.join(base_path, category, 'images'), + exist_ok=True) + os.makedirs(os.path.join(base_path, category, 'labels'), + exist_ok=True) + + def write_images(self, base_path, categories): + """ + Write dataset images + """ + for category in categories: + dest_dir = os.path.join(base_path, category, 'images') + for src_path in self.df.loc[self.df[category], 'path']: + shutil.copy2(src_path, dest_dir) + + def write_labels(self, base_path, categories, col, condition=None): + """ + Write dataset labels + """ + for category in categories: + for idx, row in self.df[self.df[category]].iterrows(): + if condition is None or condition[idx]: + dest_name = os.path.splitext(row['filename'])[0] + '.txt' + dest_path = os.path.join(base_path, category, + 'labels', dest_name) + dest_file = open(dest_path, 'w') + print(row[col], file=dest_file) + dest_file.close() + + def write_image_lists(self, base_path, categories, condition): + """ + Write image lists, using only images that meet condition + """ + for category in categories: + list_content = './' + category + '/images/' + self.df.loc[ + self.df[category] & condition, 'filename'] + list_content.to_csv(os.path.join(base_path, category + '.txt'), + header=False, index=False) + + def add_class_labels(self, detector_classes, condition=None, + input_col='auto_bbox_desc', + output_col='auto_bboxclass_desc'): + """ + Given localizer bounding boxes (which are all of class zero), + generate detector bounding boxes (which indicate class of plane) + """ + detector_classes_dict = {x:y for y, x in enumerate(detector_classes)} + self.df[output_col] = '' + for idx, row in self.df.iterrows(): + if condition is None or condition[idx]: + dest_cont = str(detector_classes_dict[row['mm']]) + row[input_col][1:].replace('\n0', '\n' + str(detector_classes_dict[row['mm']])) # Assumes input class is zero + self.df.loc[idx, output_col] = dest_cont + + def train(self, + weights='yolov7-tiny.pt', + cfg='cfg/training/yolov7-tiny.yaml', + data=None, #e.g., /code/localizer.yaml + hyp='data/hyp.scratch.tiny.yaml', + epochs=None, #e.g., 500 + batchsize=4, + workers=4, + name=None, #e.g., yolov7_localizer_01_1 + ): + """ + Train YOLOv7 model + """ + cmd = 'python train.py' + cmd += ' --weights ' + weights + cmd += ' --cfg ' + cfg + cmd += ' --data ' + data + cmd += ' --hyp ' + hyp + cmd += ' --epochs ' + str(epochs) + cmd += ' --batch-size ' + str(batchsize) + cmd += ' --workers ' + str(workers) + cmd += ' --name ' + name + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True, bufsize=1) + for line in p.stdout: + print(line) + + def test(self, + train_name, #e.g., yolov7_localizer_01_1 + test_name, #e.g., yolov7_localizer_01_1 + data, #e.g., /code/localizer.yaml + batchsize=32, + confthres=0.25, + iouthres=0.5, + task='test', + ): + """ + Run YOLOv7 model to generate labels on categories listed in yaml file + """ + weights = os.path.join('runs/train/', train_name, 'weights/best.pt') + cmd = 'python test2.py' + cmd += ' --weights ' + weights + cmd += ' --data ' + data + cmd += ' --batch-size ' + str(batchsize) + cmd += ' --conf-thres ' + str(confthres) + cmd += ' --iou-thres ' + str(iouthres) + cmd += ' --task ' + task + cmd += ' --name ' + test_name + cmd += ' --save-json' + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True, bufsize=1) + for line in p.stdout: + print(line) + + def detect(self, + train_name, # e.g., yolov7_localizer_01_1 + detect_name, # e.g., yolov7_localizer_01_1 + source, # e.g., path/filename.jpg + imgsize=640, + confthres=0.25 + ): + """ + Run YOLOv7 model on specified file(s) + """ + weights = os.path.join('runs/train/', train_name, 'weights/best.pt') + cmd = 'python detect.py' + cmd += ' --weights ' + weights + cmd += ' --source "' + source + '"' + cmd += ' --img-size ' + str(imgsize) + cmd += ' --conf-thres ' + str(confthres) + cmd += ' --save-txt' + cmd += ' --save-conf' + cmd += ' --agnostic-nms' + cmd += ' --name ' + detect_name + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True, bufsize=1) + for line in p.stdout: + print(line) + + def sort(self, + sourcedir, + planedir, + noplanedir, + logdir, + weights='../code/localizer.pt', + imgsize=640, + confthres=0.25, + device='cpu', + savejson=False, + ): + """ + Runs daemon to sort folder contents into plane and noplane folders + Enter Ctrl+c to end (otherwise, continues indefinitely) + """ + cmd = 'python sort.py' + cmd += ' --weights ' + weights + cmd += ' --source-dir ' + sourcedir + cmd += ' --plane-dir ' + planedir + cmd += ' --noplane-dir ' + noplanedir + cmd += ' --log-dir ' + logdir + cmd += ' --img-size ' + str(imgsize) + cmd += ' --conf-thres ' + str(confthres) + cmd += ' --device ' + device + cmd += ' --nosave' + cmd += ' --agnostic-nms' + if savejson: + cmd += ' --save-json' + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True) + for line in p.stdout: + print(line) + + def sort_tflite(self, + sourcedir, + planedir, + noplanedir, + logdir, + weights='../data/tflite/localizer.tflite', + savejson=False, + ): + """ + Runs daemon to sort folder contents into plane and noplane folders + using the Tensorflow Lite version of the model. + Enter Ctrl+c to end (otherwise, continues indefinitely). + """ + cmd = 'python sort_tflite.py' + cmd += ' --weights ' + weights + cmd += ' --source-dir ' + sourcedir + cmd += ' --plane-dir ' + planedir + cmd += ' --noplane-dir ' + noplanedir + cmd += ' --log-dir ' + logdir + if savejson: + cmd += ' --save-json' + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True) + for line in p.stdout: + print(line) + + def visualize_train(self, name, batch): + """ + Show some graphics generated during model training + Currently, just prints paths to graphics + """ + results_dir = os.path.join('runs/train/', name) + file_names = ['results.png', + 'test_batch' + str(batch) + '_labels.jpg', + 'test_batch' + str(batch) + '_pred.jpg'] + for file_name in file_names: + file_path = os.path.join(results_dir, file_name) + print(file_path) + + def visualize_detect(self, name, image_path): + """ + Visualize and print detector inference example + Currently, just prints path to graphic + """ + results_dir = os.path.join('runs/detect', name) + idstring = os.path.splitext(os.path.split(image_path)[1])[0] + image_path = os.path.join(results_dir, idstring + '.jpg') + text_path = os.path.join(results_dir, 'labels', idstring + '.txt') + + file = open(text_path) + print(file.readlines()) + file.close() + print(image_path) + + def check_planes_per_image(self, flag_col='auto_bbox_flag', + count_col='auto_bbox_count'): + """ + Count images with various numbers of planes found + """ + print('0 Planes:', + (self.df[flag_col] & (self.df[count_col] == 0)).sum()) + print('1 Plane :', + (self.df[flag_col] & (self.df[count_col] == 1)).sum()) + print('2+ Planes:', + (self.df[flag_col] & (self.df[count_col] >= 2)).sum()) + + def check_top_global_mm_strings(self, database_path, count=25): + """ + Print the most common unsanitized make+model strings in the + aircraft database file + """ + database_dataframe = pd.read_csv(database_path, usecols=['icao24', 'manufacturername', 'model']) + database_dataframe['string'] = database_dataframe['manufacturername'] + ' ' + database_dataframe['model'] + database_dataframe = database_dataframe[['icao24', 'string']] + database_dataframe = database_dataframe.groupby('string', as_index=False).count().sort_values(by='icao24', ascending=False) + database_dataframe = database_dataframe.reset_index(drop=True) + with pd.option_context('display.max_rows', None, 'display.max_columns', None, 'display.width', display_width): + print(database_dataframe.iloc[:count]) + + def check_confusion_matrix(self, cmpath, save=False, names=()): + """ + Show a confusion matrix previously saved to disk + """ + cm = ConfusionMatrix(0) + cm.read(cmpath) + print(cm.matrix) + print() + with np.printoptions(precision=3): + print(cm.norm(0).matrix) + print() + print(cm.norm(1).matrix) + if save: + savedir = os.path.split(cmpath)[0] + cm.plot(savedir, names, normalize=False, transpose=False, + round_to_int=True, + filename='confusion_matrix_counts') + cm.norm(0).plot(savedir, names, normalize=False, transpose=False, + filename='confusion_matrix_bytrue') + cm.norm(1).plot(savedir, names, normalize=False, transpose=False, + filename='confusion_matrix_bypred') + + def torch2onnx(self, + weights, # e.g., 'runs/train/xyz/yolov7_localizer_01_01/weights/best.pt' + imgsize=640, + confthres=0.25, + iouthres=0.5, + nms=True + ): + """ + Convert PyTorch YOLOv7 model to ONNX + """ + cmd = 'python export.py' + cmd += ' --weights ' + weights + cmd += ' --grid' + if nms: + cmd += ' --end2end' + cmd += ' --simplify' + cmd += ' --topk-all 100' + cmd += ' --iou-thres ' + str(iouthres) + cmd += ' --conf-thres ' + str(confthres) + cmd += ' --img-size ' + str(imgsize) + ' ' + str(imgsize) + cmd += ' --max-wh ' + str(imgsize) + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True) + for line in p.stdout: + print(line) + + def onnx2tf(self, + input_model, # e.g., 'runs/train/xyz/yolov7_localizer_01_01/weights/best.onnx' + output_model # e.g., '../data/tf/localizer' + ): + """ + Convert ONNX YOLOv7 model to TensorFlow + """ + cmd = 'onnx-tf convert' + cmd += ' -i ' + input_model + cmd += ' -o ' + output_model + print(cmd) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True) + for line in p.stdout: + print(line) + + def tf2tflite(self, + input_model, # e.g., '../data/tf/localizer' + output_model # e.g., '../data/tflite/localizer.tflite' + ): + """ + Convert TensorFlow YOLOv7 model to TensorFlow Lite + """ + import tensorflow as tf + tflite_model = tf.lite.TFLiteConverter\ + .from_saved_model(input_model)\ + .convert() + with open(output_model, 'wb') as output_file: + output_file.write(tflite_model) + + +def main_sequence(): + """ + Original model-building sequence, using six classes + and previously-acquired close-view data + Note: Methods train, test, detect, sort, and write_* write output to disk + """ + # Images + ds = Dataset() + ds.create_from_images('../data/2021/skyscan-datasets/dataset/close-view') + ds.load_makemodel_string('../data/databases/aircraftDatabase.csv') + ch = ds.load_makemodel_mm('../code/taxon.json', True) + ds.print_classes() + detector_classes = ['Airbus A319', 'Airbus A320', 'Airbus A321', + 'Boeing 737', 'Bombardier CL-600', 'Embraer ERJ-170'] + ds.add_usable(detector_classes, ch['helicopters']) + ds.print(title='Initial Dataset') + + # Bounding boxes and data split + ds.write_suggested_labels('../data/label_suggestions/label_suggestions.csv') + ds.load_labelbox_labels('../data/labelbox_export/export-2022-10-09T04_14_18.580Z.json') + ds.print(title='Manual Labels', condition=ds.df['manual_bbox_flag']) + ds.split_train_test(detector_classes) + categories = ['localizer_train', 'localizer_test', 'detector_train', 'detector_test'] + ds.check_categories_nonintersecting(categories) + ds.print(title='Split') + + # Write dataset to disk + # ds.write_folders('../data/dataset', categories) + # ds.write_images('../data/dataset', categories) + # ds.write_labels('../data/dataset', categories[:2], 'manual_bbox_desc') + + # Localizer + loc_name_trn = 'yolov7_localizer_01_2' # Change before retraining + loc_name_tst = 'yolov7_localizer_01_23' # Change before retesting + loc_data = '/code/localizer.yaml' + # ds.train(data=loc_data, epochs=500, name=loc_name_trn) + ds.visualize_train(loc_name_trn, 0) + # ds.test(train_name=loc_name_trn, test_name=loc_name_tst, data=loc_data) + ds.load_yolov7_labels(name=loc_name_tst) + ds.check_planes_per_image() + ds.write_image_lists('../data/dataset', categories[2:], + ds.df['auto_bbox_count']==1) + ds.add_class_labels(detector_classes, ds.df['auto_bbox_count']==1, + 'auto_bbox_desc', 'auto_bboxclass_desc') + # ds.write_labels('../data/dataset', categories[2:], 'auto_bboxclass_desc', + # ds.df['auto_bbox_count']==1) + ds.print(title='Auto Labels', condition=ds.df['auto_bbox_flag']) + + # Detector + det_name_trn = 'yolov7_detector_01_4' # Change before retraining + det_name_det = 'yolov7_detector_01_2' # Change before re-detecting + det_data = '/code/detector.yaml' + # ds.train(data=det_data, epochs=50, name=det_name_trn) + ds.visualize_train(det_name_trn, 0) + example_file = '../data/skyscan-datasets/dataset/close-view/4-23-DCA/Airbus A319 112/AA0AAB_2021-04-23-14-43-05.jpg' + # ds.detect(train_name=det_name_trn, detect_name=det_name_det, + # source=example_file) + ds.visualize_detect(det_name_det, example_file) + + # Other + # Sorting daemon + # ds.sort('../data/edge_test/tosort', '../data/edge_test/plane', + # '../data/edge_test/noplane', '../data/edge_test/log') + # Common make+models + # ds.check_top_global_mm_strings('../data/databases/aircraftDatabase.csv', 100) + # Confusion matrix + det_name_tst = 'yolov7_detector_01_3' # Change before retesting + # ds.test(train_name=det_name_trn, test_name=det_name_tst, data=det_data, + # task='val') + cmpath = os.path.join('runs/test', det_name_tst, 'confusion_matrix_counts.txt') + ds.check_confusion_matrix(cmpath, save=True, names=detector_classes) + + +def daemon(): + """ + Runs sorting daemon + """ + ds = Dataset() + ds.sort('../data/edge_test/tosort', + '../data/edge_test/plane', + '../data/edge_test/noplane', + '../data/edge_test/log', + savejson=True) + + +def compare_collections(): + """ + Look at contents of collections of images + Note: The first set of paths includes many images multiple times + """ + paths = ['../data/2021/skyscan-datasets/dataset/cruising-view', + '../data/2021/skyscan-datasets/dataset/close-view', + '../data/2022/airshow', + '../data/2022/multi/2022-11-16_field-test', + '../data/2022/multi/2022-11-18--21_weekend-home', + '../data/2021/tf/dataset-export/multi_class_train_by_aircraft', + '../data/2021/tf/dataset-export/multi_class_eval_by_aircraft', + '../data/2021/322images-training', + '../data/2021/500image-train'] + paths = ['../data/2022/airshow', + '../data/2021/skyscan-datasets/dataset/close-view', + '../data/2022/multi/2022-11-16_field-test'] + # paths = ['../data/classification/test/%02d' % (classnum,) + # for classnum in range(23)] + for path in paths: + ds = Dataset() + ds.create_from_images(path) + ds.load_makemodel_string('../data/databases/aircraftDatabase.csv') + ds.load_makemodel_mm('../code/taxon.json') + ds.print_classes(title=path) + if True: + ds = Dataset() + ds.create_from_dirs(paths) + ds.load_makemodel_string('../data/databases/aircraftDatabase.csv') + ds.load_makemodel_mm('../code/taxon.json') + ds.print_classes(title='all') + print('Identified make+models:', len(ds.df['mm'].unique())) + + +def entry_search(): + """ + Print dataset items matching command line argument + """ + paths = ['../data/2022/airshow', + '../data/2021/skyscan-datasets/dataset/close-view', + '../data/2022/multi/2022-11-16_field-test'] + col = 'make' + ds = Dataset() + ds.create_from_dirs(paths) + ds.load_makemodel_string('../data/databases/aircraftDatabase.csv') + ds.load_makemodel_mm('../code/taxon.json') + print(ds.df[ds.df[col]==sys.argv[1]]) + print(ds.df[ds.df[col]==sys.argv[1]]['icao24'].unique()) + + +def revised_sequence(): + """ + Revised sequence of steps to build two models: a "localizer" to identify + bounding boxes, and a "detector" to identify bounding boxes as well as + class of each aircraft + """ + + # FILE NAMES + # Commercial air transport images for training/validation + paths_cat = ['../data/2021/skyscan-datasets/dataset/close-view', + '../data/2022/multi/2022-11-16_field-test'] + # General aviation images for training/validation + paths_ga = ['../data/2022/airshow'] + # Images for testing + paths_test = ['../data/2023/dulles/raw'] + # Path to database of aircraft from OpenSky Network + database_path = '../data/databases/aircraftDatabase.csv' + # Lookup table with hierarchy of aircraft makes, models, and descriptions + hierarchy_path = '../code/taxon.json' + # Make/models to train detector on + detector_classes = ['Airbus A319', 'Airbus A320', 'Airbus A321', + 'Beechcraft Bonanza', + 'Boeing 737', + 'Bombardier CL-600', + 'Cessna Citation', 'Cessna Skyhawk', 'Cessna Skylane', + 'Embraer ERJ-170', + 'Piper PA-28 Cherokee', 'Piper PA-32 Cherokee Six'] + # Files in which to save lists of suggested images to manually label + ls_cat = '../data/label_suggestions/label_suggestions_revised_cat.csv' + ls_ga = '../data/label_suggestions/label_suggestions_revised_ga.csv' + ls = '../data/label_suggestions/label_suggestions_revised.csv' + ls_test = '../data/label_suggestions/label_suggestions_revised_test.csv' + # Folder in which to store dataset + dataset_dir = '../data/dataset' + # Files of hand-drawn bounding boxes, exported from LabelBox + labelbox_export_trainval = '../data/labelbox_export/export-2022-12-07T07_41_59.984Z.json' + labelbox_export_test = '../data/labelbox_export/export-2023-03-02T07_38_51.807Z.json' + + # YOLOv7 OUTPUT NAMES + # Folder name for training the localizer + loc_name_trn = 'yolov7_localizer_02_01' # Change before retraining + # Folder name for testing the localizer + loc_name_tst = 'yolov7_localizer_02_04' # Change before retesting + # Folder name for training the detector + det_name_trn = 'yolov7_detector_02_01' # Change before retraining + # Folder name for testing the detector + det_name_tst = 'yolov7_detector_02_01' # Change before re-detecting + # Folder name for testing on new data + test_name = 'yolov7_test_02_14' # Change before retesting + + # Open CAT and GA collects separately, to label different fractions thereof + ds_cat = Dataset() + ds_ga = Dataset() + ds_cat.create_from_dirs(paths_cat) + ds_ga.create_from_dirs(paths_ga) + ds_cat.load_makemodel_string(database_path) + ds_ga.load_makemodel_string(database_path) + ds_cat.load_makemodel_mm(hierarchy_path) + ch = ds_ga.load_makemodel_mm(hierarchy_path) + ds_cat.print_classes(title='Field tests') + ds_ga.print_classes(title='Airshow') + ds_cat.add_usable(detector_classes, ch['helicopters']) # + ch['homebuilt']) + ds_ga.add_usable(detector_classes, ch['helicopters']) # + ch['homebuilt']) + ds_cat.write_suggested_labels( + ls_cat, max_planes_per_class=12, max_images_per_plane=2) + ds_ga.write_suggested_labels( + ls_ga, max_planes_per_class=12, max_images_per_plane=1) + subprocess.check_output("cat %s %s > %s" % (ls_cat, ls_ga, ls), shell=True) + ds = ds_cat.concat(ds_ga) + ds.print_classes(title='Combined dataset') + + # Check that suggested files to hand-label are as expected + ds_suggest = Dataset() + ds_suggest.create_from_paths_file(ls) + ds_suggest.load_makemodel_string(database_path) + ds_suggest.load_makemodel_mm(hierarchy_path) + ds_suggest.print_classes(title='Hand-labeled') + + # Data split + if not os.path.exists(labelbox_export_trainval): + raise Exception('Use LabelBox to label bounding boxes for the files in ' + ls + ' and save the exported LabelBox file to ' + labelbox_export_trainval + ' (or specify the correct path in the code), then re-run this function.') + ds.load_labelbox_labels(labelbox_export_trainval, min_area=400) + ds.print(title='Manual Labels', condition=ds.df['manual_bbox_flag']) + ds.split_train_val_shared(detector_classes, val_fraction=0.25) + ds.print_classes(title='Detector Train', condition=ds.df['detector_train']) + ds.print_classes(title='Detector Val', condition=ds.df['detector_val']) + + # Write dataset to disk + categories = ['localizer_train', 'localizer_val', + 'detector_train', 'detector_val'] + ds.write_folders(dataset_dir, categories) + ds.write_images(dataset_dir, categories) ## + ds.write_labels(dataset_dir, categories[:2], 'manual_bbox_desc') + + # Localizer + loc_data = 'localizer_rev2.yaml' + ds.train(data=loc_data, epochs=500, name=loc_name_trn) ## + ds.visualize_train(loc_name_trn, 0) + + # Test on localizer_val data, to generate localizer confusion matrix + ds.test(train_name=loc_name_trn, test_name=loc_name_tst, + data=loc_data, task='val') ## + ds.load_yolov7_labels(name=loc_name_tst, min_area=400, + always_flag=['localizer_val']) # Used? + ds.check_planes_per_image() + cmpath = os.path.join('runs/test', loc_name_tst, + 'confusion_matrix_counts.txt') + ds.check_confusion_matrix(cmpath, save=True, names=['aircraft']) + + # Test on detector data, to generate detector training data labels + ds.df.drop(['auto_bbox_flag', 'auto_bbox_count', 'auto_bbox_desc'], + axis='columns') + ds.test(train_name=loc_name_trn, test_name=loc_name_tst+'b', + data=loc_data) ## + ds.load_yolov7_labels(name=loc_name_tst+'b', min_area=400, + always_flag=['detector_train', 'detector_val']) + ds.check_planes_per_image() + ds.write_image_lists(dataset_dir, categories[2:], + ds.df['auto_bbox_count']==1) + ds.add_class_labels(detector_classes, ds.df['auto_bbox_count']==1, + 'auto_bbox_desc', 'auto_bboxclass_desc') + ds.write_labels(dataset_dir, categories[2:], 'auto_bboxclass_desc', + ds.df['auto_bbox_count']==1) ## + + # Detector + det_data = 'detector_rev2.yaml' + ds.train(data=det_data, epochs=50, name=det_name_trn) ## + ds.visualize_train(det_name_trn, 0) + + # Test on detector_val data, to generate detector confusion matrix + ds.test(train_name=det_name_trn, test_name=det_name_tst, + data=det_data, task='val') ## + ds.load_yolov7_labels(name=det_name_tst, min_area=400, + always_flag=['detector_val']) + ds.check_planes_per_image() + cmpath = os.path.join('runs/test', det_name_tst, + 'confusion_matrix_counts.txt') + ds.check_confusion_matrix(cmpath, save=True, names=detector_classes) + + # Testing the model on a new dataset + ds = Dataset() + ds.create_from_dirs(paths_test) + ds.load_makemodel_string(database_path) + ch = ds.load_makemodel_mm(hierarchy_path) + ds.add_usable(detector_classes, ch['helicopters']) + ds.print_classes(title='Testing Data') + # Manual bounding boxes + ds.write_suggested_labels_prop(ls_test, num_planes=100, + max_images_per_plane=1) + if not os.path.exists(labelbox_export_trainval): + raise Exception('Use LabelBox to label bounding boxes for the files in ' + ls_test + ' and save the exported LabelBox file to ' + labelbox_export_test + ' (or specify the correct path in the code), then re-run this function.') + ds.load_labelbox_labels(labelbox_export_test, min_area=400) + # Write imagery and manual bounding boxes (for test of localizer) + categories = ['localizer_test', 'detector_test'] + ds.df['localizer_test'] = ds.df['manual_bbox_flag'] + ds.df['detector_test'] = ds.df['detector_usable'] + ds.write_folders(dataset_dir, categories) + ds.write_images(dataset_dir, categories) ## + ds.write_labels(dataset_dir, categories[:1], 'manual_bbox_desc') + # Generate and write auto bounding boxes (for test of detector) + ds.test(train_name=loc_name_trn, test_name=test_name+'a', + data=loc_data, task='test3') ## + ds.load_yolov7_labels(name=test_name+'a', min_area=400, + always_flag=['detector_test']) + ds.check_planes_per_image() + ds.write_image_lists(dataset_dir, categories[1:], + ds.df['auto_bbox_count']==1) + ds.add_class_labels(detector_classes, ds.df['auto_bbox_count']==1, + 'auto_bbox_desc', 'auto_bboxclass_desc') + ds.write_labels(dataset_dir, categories[1:], 'auto_bboxclass_desc', + ds.df['auto_bbox_count']==1) ## + ds.print(title='Test data', condition=ds.df['manual_bbox_count']>=1) + # Test of localizer with testing data + ds.test(train_name=loc_name_trn, test_name=test_name+'b', + data=loc_data, task='test2') ## + cmpath = os.path.join('runs/test', test_name+'b', + 'confusion_matrix_counts.txt') + ds.check_confusion_matrix(cmpath, save=True, names=['aircraft']) + # Test of detector with testing data + ds.test(train_name=det_name_trn, test_name=test_name+'c', + data=det_data, task='test') ## + cmpath = os.path.join('runs/test', test_name+'c', + 'confusion_matrix_counts.txt') + ds.check_confusion_matrix(cmpath, save=True, names=detector_classes) + + +def tflite_conversion(): + """ + Convert the model: + PyTorch --> ONNX --> TensorFlow --> TensorFlow Lite + """ + train_name = 'yolov7_localizer_02_01' + model_name = 'localizer' + ds = Dataset() + + torch_model = os.path.join('runs/train/', train_name, 'weights/best.pt') + onnx_model = os.path.join('runs/train/', train_name, 'weights/best.onnx') + tf_model = os.path.join('../data/model/tf', model_name) + tflite_model = os.path.join('../data/model/tflite', model_name + '.tflite') + + ds.torch2onnx(torch_model) + ds.onnx2tf(onnx_model, tf_model) + ds.tf2tflite(tf_model, tflite_model) + + +def daemon_tflite(): + """ + Runs TFLite version of the sorting daemon + """ + ds = Dataset() + ds.sort_tflite('../data/edge_test/tosort', + '../data/edge_test/plane', + '../data/edge_test/noplane', + '../data/edge_test/log', + savejson=True) + + +if __name__ == '__main__': + #main_sequence() + #daemon() + #compare_collections() + #entry_search() + revised_sequence() + #tflite_conversion() + #daemon_tflite() diff --git a/ai/taxon.json b/ai/taxon.json new file mode 100644 index 0000000..46127e8 --- /dev/null +++ b/ai/taxon.json @@ -0,0 +1,902 @@ +{ + "version": 0.4, + "helicopters": ["Bell", "Eurocopter", "MBB", "Robinson", "Sikorsky"], + "homebuilt": ["Van's", "(Kit)"], + "make_model_strings": { + "Aeronca": { + "Champion": [ + "Aeronca 7AC", + "Bellanca 7ACA" + ] + }, + "Aero Commander": { + "500 Family": [ + "Aero Commander 560-A" + ] + }, + "Aero Vodochody": { + "L-39 Albatros": [ + "Aero L-39 ALBATROS", + "Aero Vodochody L-39", + "Aero Vodochody L-39 C", + "Aero Vodochody L39C" + ] + }, + "Airbus": { + "A300": [ + "Airbus A300B4-662R", + "Airbus A300 B4-622R" + ], + "A319": [ + "Airbus A319 112", + "Airbus A319-112", + "Airbus A319-114", + "Airbus A319-115", + "Airbus A319 115SL", + "Airbus A319-132", + "Airbus Industrie A319-112", + "Airbus Industrie A319-114", + "Airbus Industrie A319-131", + "Airbus Industrie A319-132" + ], + "A320": [ + "Airbus A320 211", + "Airbus A320-211", + "Airbus A320 214", + "Airbus A320-214", + "Airbus A320 214SL", + "Airbus A320 232", + "Airbus A320-232", + "Airbus A320 232SL", + "Airbus A320-232SL", + "Airbus A320-251N", + "Airbus Industrie A320-211", + "Airbus Industrie A320-212", + "Airbus Industrie A320-214", + "Airbus Industrie A320-232", + "Airbus Sas A320-251N" + ], + "A321": [ + "Airbus A321-211", + "Airbus A321 211SL", + "Airbus A321-231", + "Airbus A321 231SL", + "Airbus A321-253N", + "Airbus A321-253NX", + "Airbus A321-271NX", + "Airbus Industrie A321-211", + "Airbus Industrie A321-211 (Airbus)", + "Airbus Industrie A321-231", + "Airbus Sas A321-211", + "Airbus Sas A321-213" + ], + "A330": [ + "Airbus A330 203" + ], + "A340": [ + "Airbus A340-313" + ], + "A350": [ + "Airbus Industrie A350-1041", + "Airbus Airbus A350 941" + ] + }, + "Beechcraft": { + "Model 18": [ + "Beech Beech 18 C / S" + ], + "Musketeer": [ + "Beech B24R" + ], + "Bonanza": [ + "Beech F33A", + "Beech 35-C33", + "Beech D35", + "Beech F35", + "Beech J35", + "Beech K35", + "Beech P35", + "Beech S35", + "Beech V35", + "Beech V35B", + "Beech A36", + "Raytheon Aircraft Company G36" + ], + "Model 45": [ + "Beech A45" + ], + "Baron": [ + "Beech E-55", + "Beech 58", + "Beech 95-55", + "Beech 95-B55 (T42A)", + "Hawker Beechcraft Corp G58", + "Raytheon Aircraft Company 58", + "Textron Aviation Inc G58" + ], + "Model 99": [ + "Beech 99" + ], + "Super King Air": [ + "Beech 200", + "Beech B200", + "Beech B300", + "Hawker Beechcraft Corp B300C", + "Beech C-12C Huron", + "Raytheon Aircraft Company MC-12S Liberty", + "Textron Aviation Inc B300" + ], + "Premier I": [ + "Raytheon Aircraft Company 390" + ] + }, + "Bell": { + "206": [ + "Bell 206B", + "Bell 206L-3" + ], + "407": [ + "Bell 407" + ], + "429": [ + "Bell Helicopter Textron Canada 429" + ] + }, + "Bellanca": { + "Decathlon": [ + "Bellanca 8KCAB", + "American Champion Aircraft Corp. 8KCAB (Bellanca)" + ] + }, + "Boeing": { + "717": [ + "Boeing 717 2BD", + "Boeing 717-200", + "Boeing Boeing 717-2BD" + ], + "737": [ + "Boeing 737-4B7", + "Boeing 737-71Q", + "Boeing 737-724", + "Boeing 737-73V", + "Boeing 737-752", + "Boeing 737-76Q", + "Boeing 737-7AD", + "Boeing 737-7BD", + "Boeing 737-7CT", + "Boeing 737-7H4", + "Boeing 737-7L9", + "Boeing 737-7Q8", + "Boeing 737-8", + "Boeing 737-800", + "Boeing 737-823", + "Boeing 737-824", + "Boeing 737-832", + "Boeing 737-852", + "Boeing 737-8H4", + "Boeing 737-8KN", + "Boeing 737-8Q8", + "Boeing 737-890", + "Boeing 737-9", + "Boeing 737-900ER", + "Boeing 737-924ER", + "Boeing 737-932ER", + "Boeing 737NG 7H4", + "Boeing 737NG 7H4/W", + "Boeing 737NG 800/W", + "Boeing 737NG 823", + "Boeing 737NG 823/W", + "Boeing 737NG 824/W", + "Boeing 737NG 852", + "Boeing 737NG 86N", + "Boeing 737NG 86N/W", + "Boeing 737NG 8Q8/W", + "Boeing 737NG 900ER/W", + "Boeing 737NG 924ER", + "Boeing 737NG 924ER/W", + "Boeing Boeing 737-823", + "The Boeing Company 737-8Q8 (Boeing)", + "Boeing Boeing C-40A", + "Boeing C-40C" + ], + "747": [ + "Boeing 747 4R7F", + "Boeing 747 830" + ], + "757": [ + "Boeing 757-224", + "Boeing 757-232", + "Boeing 757-251", + "Boeing 757 2Q8/W" + ], + "767": [ + "Boeing 767-223", + "Boeing 767-281", + "Boeing 767-322", + "Boeing 767-424ER", + "The Boeing Company Commercial Airplane Division 767-33A (Boeing)" + ], + "777": [ + "Boeing 777-222", + "Boeing 777-223", + "Boeing 777-300ER", + "Boeing 777 368ER", + "Boeing 777 3F2ER", + "Boeing 777 FDZ", + "Boeing Boeing 777-328(ER)" + ], + "787": [ + "Boeing 787 8", + "Boeing 787-8", + "Boeing 787 9", + "Boeing Company BOEING 787-9 Dreamliner" + ], + "B-17 Flying Fortress": [ + "Boeing B-17G" + ], + "C-17 Globemaster III": [ + "Boeing C-17A Globemaster III" + ] + }, + "Bombardier": { + "BD-100": [ + "Bombardier Challenger 300", + "Bombardier Inc BD-100-1A10" + ], + "BD-500": [ + "Airbus Canada Lp BD-500-1A11", + "Airbus Canada Ltd Ptnrsp BD-500-1A10", + "C Series Aircraft Ltd Ptnrsp BD-500-1A10" + ], + "BD-700": [ + "Bombardier Inc BD-700-1A10", + "Bombardier Inc BD-700-1A11", + "Bombardier Inc BD-700-2A12" + ], + "CL-600": [ + "Bombardier Inc CL-600-2B16", + "Bombardier Inc CL-600-2B19", + "Bombardier Inc CL-600-2C10", + "Bombardier Inc CL-600-2C11", + "Bombardier Inc CL-600-2D24", + "Bombardier Inc. CL-600-2B19 (Series 100) (Bombardier)", + "Canadair Ltd CL-600-2B16", + "Bombardier CRJ 900 LR NG", + "Bombardier Global 6000" + ] + }, + "British Aerospace": { + "Jetstream 3101": [ + "British Aerospace BAE JETSTREAM 3101" + ] + }, + "Cessna": { + "120/140": [ + "Cessna 120", + "Cessna 140" + ], + "150/152": [ + "Cessna 150F", + "Cessna 150G", + "Cessna 150H", + "Cessna 150J", + "Cessna 150L", + "Cessna 150M", + "Cessna 152" + ], + "170": [ + "Cessna 170B" + ], + "Skyhawk": [ + "Cessna 172", + "Cessna 172D", + "Cessna 172E", + "Cessna 172F", + "Cessna 172G", + "Cessna 172H", + "Cessna 172K", + "Cessna 172L", + "Cessna 172M", + "Cessna 172N", + "Cessna 172P", + "Cessna 172R", + "Cessna 172S", + "Cessna R172K", + "Cessna Aircraft Company 172M (Cessna)" + ], + "Skylark": [ + "Cessna 175" + ], + "Cardinal": [ + "Cessna 177A", + "Cessna 177B", + "Cessna 177RG" + ], + "Skywagon": [ + "Cessna 180", + "Cessna 180J", + "Cessna 180K", + "Cessna A185F" + ], + "Skylane": [ + "Cessna 182A", + "Cessna 182C", + "Cessna 182F", + "Cessna 182J", + "Cessna 182N", + "Cessna 182P", + "Cessna 182Q", + "Cessna 182R", + "Cessna 182S", + "Cessna 182T", + "Cessna R182", + "Cessna T182T", + "Cessna TR182", + "Cessna Aircraft Company 182Q (Cessna)" + ], + "190/195": [ + "Cessna 195", + "Cessna 195A" + ], + "206": [ + "Cessna U206G", + "Cessna T206H" + ], + "Caravan": [ + "Cessna 208", + "Cessna 208B" + ], + "Centurion": [ + "Cessna 210J", + "Cessna T210L", + "Cessna 210M", + "Cessna T210N" + ], + "310/320": [ + "Cessna 310E", + "Cessna 310K", + "Cessna 310R", + "Cessna 320F" + ], + "401/402/411/421": [ + "Cessna 402B", + "Cessna 421C" + ], + "SkyCourier": [ + "Textron Aviation Inc 408" + ], + "414": [ + "Cessna 414" + ], + "Citation": [ + "Cessna 501", + "Cessna 510", + "Cessna 525B", + "Cessna 550", + "Cessna 560", + "Cessna 560XL", + "Cessna 650", + "Cessna 680", + "Cessna 750", + "Cessna Aircraft Co 560XLS", + "Cessna Citation Excel", + "Cessna Citation S/II", + "Cessna Citation V", + "Cessna Citation VII", + "Cessna Citation XLS", + "Cessna UC-35C Citation", + "Textron Aviation Inc 525B", + "Textron Aviation Inc 525C", + "Textron Aviation Inc 680A" + ] + }, + "Cirrus": { + "SR20/22": [ + "Cirrus Design Corp SR20", + "Cirrus Design Corp SR22", + "Cirrus Design Corp SR22T" + ], + "SF50": [ + "Cirrus Design Corp SF50" + ] + }, + "Consolidated Vultee": { + "BT-13 Valiant": [ + "Consolidated Vultee BT-13A" + ], + "PBY Catalina": [ + "Consolidated Vultee 28-5ACF" + ] + }, + "Curtiss": { + "P-40 Warhawk": [ + "Curtiss Wright P-40N" + ] + }, + "Dassault": { + "Falcon": [ + "Dassault FALCON 900 EX", + "Dassault Aviation FALCON 7X", + "Dassault Aviation FALCON 900EX" + ] + }, + "De Havilland": { + "Dash 8": [ + "Bombardier Inc. DHC-8-402 (Dehavilland)" + ] + }, + "Diamond": { + "DA40": [ + "Diamond Aircraft Ind Inc DA 40" + ] + }, + "Douglas": { + "DC-3": [ + "Douglas DC3A", + "Douglas DC3C-R", + "Douglas DC3C-S1C3G", + "Douglas DC3C-S4C4G" + ] + }, + "Embraer": { + "ERJ-135/140/145": [ + "Embraer EMB-135LR", + "Embraer ERJ-145 LR", + "Embraer EMB-145LR", + "Embraer EMB-145XR" + ], + "ERJ-170": [ + "Embraer ERJ 170-100 LR", + "Embraer ERJ 170-100 SE", + "Embraer ERJ 170-100SU", + "Embraer ERJ 170-200 LR", + "Embraer S A ERJ 170-200 LR", + "Embraer-empresa Brasileira De ERJ 170-200 LR", + "Yabora Industria Aeronautica S ERJ 170-200 LR" + ], + "ERJ-175": [ + "Embraer EMB-175 200LR" + ], + "ERJ-190": [ + "Embraer ERJ 190-100 IGW", + "Embraer EMB-190 AR" + ], + "EMB-500": [ + "Embraer Sa EMB-500", + "Embraer-empresa Brasileira De EMB-500" + ], + "EMB-505": [ + "Embraer Phenom 300", + "Embraer S A EMB-505", + "Embraer Executive Aircraft Inc EMB-505", + "Embraer-empresa Brasileira De EMB-505" + ], + "EMB-545/550": [ + "Embraer S A EMB-545", + "Embraer Sa EMB-550" + ] + }, + "Eurocopter": { + "BK-117": [ + "Eurocopter Deutschland Gmbh MBB-BK 117 C-2" + ], + "EC-120": [ + "Eurocopter EC120B" + ], + "EC-135": [ + "Eurocopter Deutschland Gmbh EC 135 P2+" + ], + "AS-350": [ + "Eurocopter AS 350 B3" + ] + }, + "Evolution": { + "Revo": [ + "Evolution Aircraft Inc REVO" + ] + }, + "Extra Flugzeugbau": { + "EA-300": [ + "Extra Flugzeugproduktions EA 300/SC" + ] + }, + "Flight Design": { + "CT": [ + "Flight Design Gmbh CTLS" + ] + }, + "Game Composites": { + "GB1 GameBird": [ + "Game Composites Llc GB1 GAMEBIRD" + ] + }, + "Grumman": { + "F6F Hellcat": [ + "Grumman A/c Eng Corp-nichols F6F-5" + ] + }, + "Grumman American": { + "AA-5": [ + "Grumman American Avn. Corp. AA-5B" + ] + }, + "Gulfstream": { + "GIV": [ + "Gulfstream Aerospace G-IV", + "Gulfstream Aerospace GIV-X (G450)" + ], + "GV": [ + "Gulfstream Aerospace GV", + "Gulfstream Aerospace G-V", + "Gulfstream Aerospace GV-SP (G550)" + ], + "GVI": [ + "Gulfstream Aerospace Corp GVI(G650ER)", + "Gulfstream Aerospace Corp GVI (G650ER)" + ] + }, + "Hawker": { + "400": [ + "Raytheon Aircraft Company 400A", + "Hawker Beechcraft Corp 400A" + ], + "800/900": [ + "British Aerospace 125 700A", + "Raytheon Aircraft Company HAWKER 800XP", + "Raytheon Aircraft Company 125 800XP", + "Hawker Beechcraft Corp Hawker 900XP", + "Hawker Beechcraft Corp HAWKER 900XP" + ] + }, + "Honda": { + "HondaJet": [ + "Honda Aircraft Co Llc HA-420" + ] + }, + "Howard": { + "DGA-15": [ + "Howard Aircraft DGA-15P" + ] + }, + "IAI": { + "G150": [ + "Iai Ltd GULFSTREAM G150" + ], + "G200": [ + "Israel Aerospace Industriesltd G200", + "Israel Aircraft Industries GULFSTREAM 200" + ], + "G280": [ + "Iai Ltd GULFSTREAM G280" + ] + }, + "Learjet": { + "31": [ + "Learjet Inc 31", + "Learjet Inc 31A", + "Bombardier Learjet 31 A" + ], + "35/36": [ + "Gates Lear Jet 36A", + "Gates Lear Jet 36 A" + ], + "45": [ + "Learjet Inc 45", + "Bombardier Learjet 45" + ], + "60": [ + "Learjet Inc 60" + ] + }, + "Lockheed": { + "Hercules": [ + "Lockheed L-100-30 Hercules" + ], + "Silver Star": [ + "Canadair T33-AN MK3" + ] + }, + "Luscombe": { + "8": [ + "Luscombe 8A" + ] + }, + "Maule": { + "M-7": [ + "Maule M-7-235C" + ] + }, + "MBB": { + "Bo 105": [ + "Mbb BO105 CBS4" + ] + }, + "Mooney": { + "M20": [ + "Mooney M20C", + "Mooney M20E", + "Mooney M20F", + "Mooney M20J", + "Mooney M20M", + "Mooney M20R", + "Mooney Aircraft Corp. M20K", + "Mooney Airplane Co Inc M20TN" + ] + }, + "North American": { + "B-25 Mitchell": [ + "North American B-25H", + "North American B-25J" + ], + "P-51 Mustang": [ + "North American F-51", + "North American F-51D", + "North American/aero Classics P-51D" + ], + "T-6 Texan": [ + "North American AT-6D Texan", + "North American T-6G", + "North American T-6G Texan", + "North American HARVARD 2", + "North American SNJ-2", + "North American SNJ-6" + ], + "T-28 Trojan": [ + "North American T-28B", + "North American NA-200" + ] + }, + "Northrop": { + "F-5": [ + "Northrop RF-5A Freedom Fighter" + ] + }, + "Pilatus": { + "PC-12": [ + "Pilatus PC-12", + "Pilatus PC-12/45", + "Pilatus PC-XII 45", + "Pilatus Aircraft Ltd PC-12", + "Pilatus Aircraft Ltd PC-12/47E" + ] + }, + "Piper": { + "J-3 Cub": [ + "Piper J3C-65" + ], + "PA-12 Super Cruiser": [ + "Piper PA-12" + ], + "PA-18 Super Cub": [ + "Piper PA-18-150" + ], + "PA-20/22": [ + "Piper PA-22-108", + "Piper PA-22-150" + ], + "PA-23 Aztec": [ + "Piper PA-23-250" + ], + "PA-24 Comanche": [ + "Piper PA-24", + "Piper PA-24-250", + "Piper PA-24-260", + "Piper PA-24-400" + ], + "PA-28 Cherokee": [ + "Piper PA-28-140", + "Piper PA-28-151", + "Piper PA-28-161", + "Piper PA-28-180", + "Piper PA-28R-180", + "Piper PA-28-181", + "Piper PA-28R-200", + "Piper PA-28RT-201T", + "Piper PA-28-235", + "Piper PA-28-236", + "Piper Aircraft Corporation PA-28-140 (Piper)" + ], + "PA-30 Twin Comanche": [ + "Piper PA-30" + ], + "PA-31 Navajo": [ + "Piper PA-31T1", + "Piper PA-31T1 Cheyenne I", + "Piper PA-31-350" + ], + "PA-32 Cherokee Six": [ + "Piper PA-32-260", + "Piper PA-32-300", + "Piper PA-32R-300", + "Piper PA-32R-301", + "Piper PA-32R-301T", + "Piper PA-32RT-300" + ], + "PA-34 Seneca": [ + "Piper PA-34-200T" + ], + "PA-46": [ + "Piper PA46-500TP", + "Piper Aircraft Inc PA46-500TP", + "Piper Aircraft Inc PA-46-600TP" + ] + }, + "Quest": { + "Kodiak": [ + "Quest Aircraft Co Inc KODIAK 100" + ] + }, + "Robinson": { + "R22": [ + "Robinson R22 Beta II", + "Robinson Helicopter R22 BETA" + ], + "R44": [ + "Robinson R44 Raven II", + "Robinson Helicopter Company R44 II" + ], + "R66": [ + "Robinson Helicopter Co R66" + ] + }, + "Rockwell": { + "OV-10 Bronco": [ + "Rockwell OV-10D" + ] + }, + "Scottish Aviation": { + "Bulldog": [ + "Scottish Aviation SERIES 100 MDL 101" + ] + }, + "Siai-Marchetti": { + "S.211": [ + "Siai Marchetti S-211", + "Siai-marchetti Srl S 211" + ], + "SF.260": [ + "Siai-marchetti SF260" + ] + }, + "Sikorsky": { + "S-76": [ + "Sikorsky S-76B" + ] + }, + "Socata": { + "TB": [ + "Socata TB 30 EPSILON", + "Compagnie Daher TB 30 EPSILON" + ], + "TBM": [ + "Socata TBM 700", + "Compagnie Daher TBM 700", + "Eads Socata TBM 850" + ] + }, + "Stearman": { + "Model 75": [ + "Boeing A75N1(PT17)" + ] + }, + "Taylorcraft": { + "B": [ + "Taylorcraft BC12-D" + ] + }, + "Van's": { + "RV-4": [ + "Bendure Ryan RV-4", + "Giffin Walter C RV-4", + "Peterson Keith D HARMON ROCKET" + ], + "RV-6": [ + "Barbee Michael C EXPERMINTAL RV-6", + "Cole Stanley T RV-6", + "Dane Julie RV6A", + "Kormylo Gene J VANS RV6A", + "Myers R.w. RV-6", + "Reisinger James A RV-6", + "Robb David M VANS RV6" + ], + "RV-7": [ + "Gary A Barnes RV-7A", + "James K Hamilton RV-7" + ], + "RV-8": [ + "Clark Michael J VANS RV-8", + "David M Schmitz/derek C James RV-8", + "Hammes Charles J VANS RV8", + "Keeling Larry RV8", + "Paine Lauran Jr RV-8", + "Quest One Llc RV-8", + "Richcreek Karl J VANS RV8", + "Rv83137 Llc RV-8" + ], + "RV-9": [ + "Nyberg Randall L RV-9" + ], + "RV-10": [ + "Hecker Robert J VANS RV-10", + "Paul Jeffrey Schwalen VANS RV10", + "Reeves Christopher D VANS RV10", + "Richard Dupee VANS RV-10", + "Sherman Michael VANS RV-10", + "Van Thomme Brendon S RV-10" + ], + "RV-12": [ + "Steven G Isaacs RV-12", + "Vans Aircraft Inc RV-12" + ], + "RV-14": [ + "Bernard L Hartnell RV 14A", + "Eisele Donald L VANS RV-14", + "Howard Dean RV-14", + "Meyers Robert A RV-14A", + "Michael J Konz RV-14A" + ] + }, + "Vought": { + "F4U Corsair": [ + "Chance Vought F4U-5", + "Goodyear FG1D" + ] + }, + "(Kit)": { + "Aeroprakt A-22": [ + "Aeroprakt Manufacturing Sp Zoo A22LS" + ], + "AMD Zodiac": [ + "Jaime Rabins ZODIAC 650 B" + ], + "Barrows Bearhawk": [ + "Yates Jared N AVIPRO BEARHAWK" + ], + "Bede BD-4": [ + "Mahoney Steven BD-4" + ], + "Cub Crafters CC-11": [ + "Cubcrafters Inc CC11-160" + ], + "Cub Crafters CC-19": [ + "Cub Crafters Inc CC19-180" + ], + "Giles G-202": [ + "Bergevin Steven GILES G-202" + ], + "Glasair III": [ + "Broen Peter C GLASAIR III", + "Moman Guy E Jr GLASAIR III" + ], + "Glasair Sportsman 2+2": [ + "Moring Ronald L SPORTSMAN 2+2" + ], + "Sling TSi": [ + "Christopher P Bloch SLING 4 TSI" + ], + "Just Highlander": [ + "Kyle Bessent JUST ACFT HIGHLANDER", + "Marlowe Thooft JUST ACFT HIGHLANDER" + ], + "Lancair Model 42": [ + "Lancair Company LC42-550FG" + ], + "Lockwood Aircam": [ + "Red Eagle Aviation Llc AIRCAM" + ], + "Quickie Q2": [ + "Scheevel Jay QUICKIE TRI-Q2" + ], + "Rans S-9": [ + "Sailer Denis R D&S-9A" + ], + "Robinson Special": [ + "Lush Michael ROBINSON SPECIAL" + ], + "Velocity SE": [ + "Hancock William J VELOCITY-STANDARD", + "Martino Thomas G VELCITY 173 ELITE RG" + ] + } + } +} diff --git a/ai/test2.py b/ai/test2.py new file mode 100644 index 0000000..965cbe8 --- /dev/null +++ b/ai/test2.py @@ -0,0 +1,354 @@ +import argparse +import json +import os +from pathlib import Path +from threading import Thread + +import numpy as np +import torch +import yaml +from tqdm import tqdm + +from models.experimental import attempt_load +from utils.datasets import create_dataloader +from utils.general import coco80_to_coco91_class, check_dataset, check_file, check_img_size, check_requirements, \ + box_iou, non_max_suppression, scale_coords, xyxy2xywh, xywh2xyxy, set_logging, increment_path, colorstr +from utils.metrics2 import ap_per_class, ConfusionMatrix +from utils.plots import plot_images, output_to_target, plot_study_txt +from utils.torch_utils import select_device, time_synchronized, TracedModel + + +def test(data, + weights=None, + batch_size=32, + imgsz=640, + conf_thres=0.001, + iou_thres=0.6, # for NMS + save_json=False, + single_cls=False, + augment=False, + verbose=False, + model=None, + dataloader=None, + save_dir=Path(''), # for saving images + save_txt=False, # for auto-labelling + save_hybrid=False, # for hybrid auto-labelling + save_conf=False, # save auto-label confidences + plots=True, + wandb_logger=None, + compute_loss=None, + half_precision=True, + trace=False, + is_coco=False, + v5_metric=False): + # Initialize/load model and set device + training = model is not None + if training: # called by train.py + device = next(model.parameters()).device # get model device + + else: # called directly + set_logging() + device = select_device(opt.device, batch_size=batch_size) + + # Directories + save_dir = Path(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run + (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir + + # Load model + model = attempt_load(weights, map_location=device) # load FP32 model + gs = max(int(model.stride.max()), 32) # grid size (max stride) + imgsz = check_img_size(imgsz, s=gs) # check img_size + + if trace: + model = TracedModel(model, device, imgsz) + + # Half + half = device.type != 'cpu' and half_precision # half precision only supported on CUDA + if half: + model.half() + + # Configure + model.eval() + if isinstance(data, str): + is_coco = data.endswith('coco.yaml') + with open(data) as f: + data = yaml.load(f, Loader=yaml.SafeLoader) + check_dataset(data) # check + nc = 1 if single_cls else int(data['nc']) # number of classes + iouv = torch.linspace(0.5, 0.95, 10).to(device) # iou vector for mAP@0.5:0.95 + niou = iouv.numel() + + # Logging + log_imgs = 0 + if wandb_logger and wandb_logger.wandb: + log_imgs = min(wandb_logger.log_imgs, 100) + # Dataloader + if not training: + if device.type != 'cpu': + model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters()))) # run once + task = opt.task if opt.task in ('train', 'val', 'test', 'test2', 'test3') else 'val' # path to train/val/test images + dataloader = create_dataloader(data[task], imgsz, batch_size, gs, opt, pad=0.5, rect=True, + prefix=colorstr(f'{task}: '))[0] + + if v5_metric: + print("Testing with YOLOv5 AP metric...") + + seen = 0 + confusion_matrix = ConfusionMatrix(nc=nc) + names = {k: v for k, v in enumerate(model.names if hasattr(model, 'names') else model.module.names)} + coco91class = coco80_to_coco91_class() + s = ('%20s' + '%12s' * 6) % ('Class', 'Images', 'Labels', 'P', 'R', 'mAP@.5', 'mAP@.5:.95') + p, r, f1, mp, mr, map50, map, t0, t1 = 0., 0., 0., 0., 0., 0., 0., 0., 0. + loss = torch.zeros(3, device=device) + jdict, stats, ap, ap_class, wandb_images = [], [], [], [], [] + for batch_i, (img, targets, paths, shapes) in enumerate(tqdm(dataloader, desc=s)): + img = img.to(device, non_blocking=True) + img = img.half() if half else img.float() # uint8 to fp16/32 + img /= 255.0 # 0 - 255 to 0.0 - 1.0 + targets = targets.to(device) + nb, _, height, width = img.shape # batch size, channels, height, width + + with torch.no_grad(): + # Run model + t = time_synchronized() + out, train_out = model(img, augment=augment) # inference and training outputs + t0 += time_synchronized() - t + + # Compute loss + if compute_loss: + loss += compute_loss([x.float() for x in train_out], targets)[1][:3] # box, obj, cls + + # Run NMS + targets[:, 2:] *= torch.Tensor([width, height, width, height]).to(device) # to pixels + lb = [targets[targets[:, 0] == i, 1:] for i in range(nb)] if save_hybrid else [] # for autolabelling + t = time_synchronized() + out = non_max_suppression(out, conf_thres=conf_thres, iou_thres=iou_thres, labels=lb, multi_label=True) + t1 += time_synchronized() - t + + # Statistics per image + for si, pred in enumerate(out): + labels = targets[targets[:, 0] == si, 1:] + nl = len(labels) + tcls = labels[:, 0].tolist() if nl else [] # target class + path = Path(paths[si]) + seen += 1 + + if len(pred) == 0: + if nl: + stats.append((torch.zeros(0, niou, dtype=torch.bool), torch.Tensor(), torch.Tensor(), tcls)) + continue + + # Predictions + predn = pred.clone() + scale_coords(img[si].shape[1:], predn[:, :4], shapes[si][0], shapes[si][1]) # native-space pred + + # Append to text file + if save_txt: + gn = torch.tensor(shapes[si][0])[[1, 0, 1, 0]] # normalization gain whwh + for *xyxy, conf, cls in predn.tolist(): + xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh + line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format + with open(save_dir / 'labels' / (path.stem + '.txt'), 'a') as f: + f.write(('%g ' * len(line)).rstrip() % line + '\n') + + # W&B logging - Media Panel Plots + if len(wandb_images) < log_imgs and wandb_logger.current_epoch > 0: # Check for test operation + if wandb_logger.current_epoch % wandb_logger.bbox_interval == 0: + box_data = [{"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]}, + "class_id": int(cls), + "box_caption": "%s %.3f" % (names[cls], conf), + "scores": {"class_score": conf}, + "domain": "pixel"} for *xyxy, conf, cls in pred.tolist()] + boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space + wandb_images.append(wandb_logger.wandb.Image(img[si], boxes=boxes, caption=path.name)) + wandb_logger.log_training_progress(predn, path, names) if wandb_logger and wandb_logger.wandb_run else None + + # Append to pycocotools JSON dictionary + if save_json: + # [{"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}, ... + image_id = int(path.stem) if path.stem.isnumeric() else path.stem + box = xyxy2xywh(predn[:, :4]) # xywh + box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner + for p, b in zip(pred.tolist(), box.tolist()): + jdict.append({'image_id': image_id, + 'category_id': coco91class[int(p[5])] if is_coco else int(p[5]), + 'bbox': [round(x, 3) for x in b], + 'score': round(p[4], 5)}) + + # Assign all predictions as incorrect + correct = torch.zeros(pred.shape[0], niou, dtype=torch.bool, device=device) + if nl: + detected = [] # target indices + tcls_tensor = labels[:, 0] + + # target boxes + tbox = xywh2xyxy(labels[:, 1:5]) + scale_coords(img[si].shape[1:], tbox, shapes[si][0], shapes[si][1]) # native-space labels + if plots: + confusion_matrix.process_batch(predn, torch.cat((labels[:, 0:1], tbox), 1)) + + # Per target class + for cls in torch.unique(tcls_tensor): + ti = (cls == tcls_tensor).nonzero(as_tuple=False).view(-1) # prediction indices + pi = (cls == pred[:, 5]).nonzero(as_tuple=False).view(-1) # target indices + + # Search for detections + if pi.shape[0]: + # Prediction to target ious + ious, i = box_iou(predn[pi, :4], tbox[ti]).max(1) # best ious, indices + + # Append detections + detected_set = set() + for j in (ious > iouv[0]).nonzero(as_tuple=False): + d = ti[i[j]] # detected target + if d.item() not in detected_set: + detected_set.add(d.item()) + detected.append(d) + correct[pi[j]] = ious[j] > iouv # iou_thres is 1xn + if len(detected) == nl: # all targets already located in image + break + + # Append statistics (correct, conf, pcls, tcls) + stats.append((correct.cpu(), pred[:, 4].cpu(), pred[:, 5].cpu(), tcls)) + + # Plot images + if plots and batch_i < 3: + f = save_dir / f'test_batch{batch_i}_labels.jpg' # labels + Thread(target=plot_images, args=(img, targets, paths, f, names), daemon=True).start() + f = save_dir / f'test_batch{batch_i}_pred.jpg' # predictions + Thread(target=plot_images, args=(img, output_to_target(out), paths, f, names), daemon=True).start() + + # Compute statistics + stats = [np.concatenate(x, 0) for x in zip(*stats)] # to numpy + if len(stats) and stats[0].any(): + p, r, ap, f1, ap_class = ap_per_class(*stats, plot=plots, v5_metric=v5_metric, save_dir=save_dir, names=names) + ap50, ap = ap[:, 0], ap.mean(1) # AP@0.5, AP@0.5:0.95 + mp, mr, map50, map = p.mean(), r.mean(), ap50.mean(), ap.mean() + nt = np.bincount(stats[3].astype(np.int64), minlength=nc) # number of targets per class + else: + nt = torch.zeros(1) + + # Print results + pf = '%20s' + '%12i' * 2 + '%12.3g' * 4 # print format + print(pf % ('all', seen, nt.sum(), mp, mr, map50, map)) + + # Print results per class + if (verbose or (nc < 50 and not training)) and nc > 1 and len(stats): + for i, c in enumerate(ap_class): + print(pf % (names[c], seen, nt[c], p[i], r[i], ap50[i], ap[i])) + + # Print speeds + t = tuple(x / seen * 1E3 for x in (t0, t1, t0 + t1)) + (imgsz, imgsz, batch_size) # tuple + if not training: + print('Speed: %.1f/%.1f/%.1f ms inference/NMS/total per %gx%g image at batch-size %g' % t) + + # Plots + if plots: + confusion_matrix.plot(save_dir=save_dir, names=list(names.values())) + confusion_matrix.write(os.path.join(save_dir, 'confusion_matrix_counts.txt')) + if wandb_logger and wandb_logger.wandb: + val_batches = [wandb_logger.wandb.Image(str(f), caption=f.name) for f in sorted(save_dir.glob('test*.jpg'))] + wandb_logger.log({"Validation": val_batches}) + if wandb_images: + wandb_logger.log({"Bounding Box Debugger/Images": wandb_images}) + + # Save JSON + if save_json and len(jdict): + w = Path(weights[0] if isinstance(weights, list) else weights).stem if weights is not None else '' # weights + anno_json = './coco/annotations/instances_val2017.json' # annotations json + pred_json = str(save_dir / f"{w}_predictions.json") # predictions json + print('\nEvaluating pycocotools mAP... saving %s...' % pred_json) + with open(pred_json, 'w') as f: + json.dump(jdict, f) + + try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb + from pycocotools.coco import COCO + from pycocotools.cocoeval import COCOeval + + anno = COCO(anno_json) # init annotations api + pred = anno.loadRes(pred_json) # init predictions api + eval = COCOeval(anno, pred, 'bbox') + if is_coco: + eval.params.imgIds = [int(Path(x).stem) for x in dataloader.dataset.img_files] # image IDs to evaluate + eval.evaluate() + eval.accumulate() + eval.summarize() + map, map50 = eval.stats[:2] # update results (mAP@0.5:0.95, mAP@0.5) + except Exception as e: + print(f'pycocotools unable to run: {e}') + + # Return results + model.float() # for training + if not training: + s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else '' + print(f"Results saved to {save_dir}{s}") + maps = np.zeros(nc) + map + for i, c in enumerate(ap_class): + maps[c] = ap[i] + return (mp, mr, map50, map, *(loss.cpu() / len(dataloader)).tolist()), maps, t + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(prog='test.py') + parser.add_argument('--weights', nargs='+', type=str, default='yolov7.pt', help='model.pt path(s)') + parser.add_argument('--data', type=str, default='data/coco.yaml', help='*.data path') + parser.add_argument('--batch-size', type=int, default=32, help='size of each image batch') + parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)') + parser.add_argument('--conf-thres', type=float, default=0.001, help='object confidence threshold') + parser.add_argument('--iou-thres', type=float, default=0.65, help='IOU threshold for NMS') + parser.add_argument('--task', default='val', help='train, val, test, speed or study') + parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu') + parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset') + parser.add_argument('--augment', action='store_true', help='augmented inference') + parser.add_argument('--verbose', action='store_true', help='report mAP by class') + parser.add_argument('--save-txt', action='store_true', help='save results to *.txt') + parser.add_argument('--save-hybrid', action='store_true', help='save label+prediction hybrid results to *.txt') + parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels') + parser.add_argument('--save-json', action='store_true', help='save a cocoapi-compatible JSON results file') + parser.add_argument('--project', default='runs/test', help='save to project/name') + parser.add_argument('--name', default='exp', help='save to project/name') + parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment') + parser.add_argument('--no-trace', action='store_true', help='don`t trace model') + parser.add_argument('--v5-metric', action='store_true', help='assume maximum recall as 1.0 in AP calculation') + opt = parser.parse_args() + opt.save_json |= opt.data.endswith('coco.yaml') + opt.data = check_file(opt.data) # check file + print(opt) + #check_requirements() + + if opt.task in ('train', 'val', 'test', 'test2', 'test3'): # run normally + test(opt.data, + opt.weights, + opt.batch_size, + opt.img_size, + opt.conf_thres, + opt.iou_thres, + opt.save_json, + opt.single_cls, + opt.augment, + opt.verbose, + save_txt=opt.save_txt | opt.save_hybrid, + save_hybrid=opt.save_hybrid, + save_conf=opt.save_conf, + trace=not opt.no_trace, + v5_metric=opt.v5_metric + ) + + elif opt.task == 'speed': # speed benchmarks + for w in opt.weights: + test(opt.data, w, opt.batch_size, opt.img_size, 0.25, 0.45, save_json=False, plots=False, v5_metric=opt.v5_metric) + + elif opt.task == 'study': # run over a range of settings and save/plot + # python test.py --task study --data coco.yaml --iou 0.65 --weights yolov7.pt + x = list(range(256, 1536 + 128, 128)) # x axis (image sizes) + for w in opt.weights: + f = f'study_{Path(opt.data).stem}_{Path(w).stem}.txt' # filename to save to + y = [] # y axis + for i in x: # img-size + print(f'\nRunning {f} point {i}...') + r, _, t = test(opt.data, w, opt.batch_size, i, opt.conf_thres, opt.iou_thres, opt.save_json, + plots=False, v5_metric=opt.v5_metric) + y.append(r + t) # results and times + np.savetxt(f, y, fmt='%10.4g') # save + os.system('zip -r study.zip study_*.txt') + plot_study_txt(x=x) # plot