TypeError while running ./docker/run_pose_estimation.bsh

[mlngan]

I successfully built a docker container that runs in --cpu-only mode following the instructions here - https://github.com/VisionSystemsInc/pix2face/tree/master/docker. During execution of the ./docker/run_pose_estimation.bsh script, the following python error occurs -- TypeError: Unable to convert function return value to a Python type! The signature was (arg0: face3d.perspective_camera_parameters) -> vgl_rotation_3d. Could there be an issue or mismatch between the python bindings?

./docker/run_pose_estimation.bsh images output --cpu-only
INPUT_DIR=images
OUTPUT_DIR=output
Running pix2face_cpu docker image
Running on CPU
loading /pix2face/pix2face_net/data/models/pix2face_unet_v10_pytorch0.4_cpu.pt ...
...done.
/input_dir/image.jpg
Traceback (most recent call last):
File "/pix2face/scripts/estimate_pose_batch.py", line 49, in
main(input_fname, output_fname)
File "/pix2face/scripts/estimate_pose_batch.py", line 35, in main
pose = pix2face_estimation.camera_estimation.estimate_head_pose(img, pix2face_net, cuda_device=cuda_device)
File "/pix2face/python/pix2face_estimation/camera_estimation.py", line 18, in estimate_head_pose
return extract_head_pose(estimate_camera(image, pix2face_net, cuda_device=cuda_device))
File "/pix2face/python/pix2face_estimation/camera_estimation.py", line 13, in extract_head_pose
yaw, pitch, roll = decompose_camera_rotation(camera_params.rotation.as_matrix(), pitch_offset=-7)
TypeError: Unable to convert function return value to a Python type! The signature was
(arg0: face3d.perspective_camera_parameters) -> vgl_rotation_3d

[scott-vsi]

Sorry about this. I pushed a fix for it (c68b635). It seems to be an issue with the necessary pybind11 overload not being loaded. So I manually loaded it. I have not audited the repo to see if this problem exists anywhere else, but if it does, you can create another issue and it should be easy to fix.

[scott-vsi]

I think we need to do this in the face3d module

  py::module importlib_util = py::module::import("importlib.util");
  if (!importlib_util.attr("find_spec")("vxl.vgl.algo").is_none()) {
    py::module::import("vxl.vgl");
  }