`requirements` file

[Nandini-Jaiswal]

Can you please provide the requirements.txt?

[snoop2head]

@Nandini-Jaiswal

Sorry for the late reply.
I have attached requirements.txt as you requested! If you have any more questions regarding the setup, please ask!

[Nandini-Jaiswal]

@Nandini-Jaiswal

Sorry for the late reply. I have attached requirements.txt as you requested! If you have any more questions regarding the setup, please ask!

Thanks so much. I have a few questions related to the datasets. You may answer if you have time.

• The Angelina dataset has the coordinates and the corresponding labels. Can we directly fit the yolo model on that?
• If we crop based on the coordinates for the Angelina dataset and train along with the Kaggle dataset for the classification purpose, will that be fine too?

Did you follow a similar approach?

Also, I don't intend to use DSBI initially, and I can't access the braille_naturals dataset (I think I need to follow some steps on their site; I will do that soon).
Will the code work without them?

Sorry for asking so many questions; I will go through the code to understand everything. I just needed a heads-up.

[snoop2head]

@Nandini-Jaiswal

• Indeed, I performed additional preprocessing by converting the labels into YOLO format.
• Yes, collecting as much braille data as possible is recommended, given how scarce it is.

I suggest downloading braille_naturals because, at the time this repository was created, it was the only dataset that included background scenes beyond the standard book pages. The code will still work without it; in that case, the model can identify braille dots on book pages only.

Thank you for your interest! If you have any questions, please feel free to ask.

[Nandini-Jaiswal]

@Nandini-Jaiswal

• Indeed, I performed additional preprocessing by converting the labels into YOLO format.
• Yes, collecting as much braille data as possible is recommended, given how scarce it is.

I suggest downloading braille_naturals because, at the time this repository was created, it was the only dataset that included background scenes beyond the standard book pages. The code will still work without it; in that case, the model can identify braille dots on book pages only.

Thank you for your interest! If you have any questions, please feel free to ask.

Sure, I will do this and give you an update ASAP. Thank you so much for your reply.

[Nandini-Jaiswal]

Hello @snoop2head, I'm downloading the datasets. After that, what should I do?

In another issue you mentioned that we need to crop the images for angelina and DSBI. So, should I run crop_bbox.py first and then train.py?

[snoop2head]

@Nandini-Jaiswal

Yes, I would do that in order to train the pseudo-label the natural braille dataset. However, if you want to train a object detector that is specialized only for the book scanning purposes, you can skip that part and train the object detector according to the yolov8 dataset format.