Question about preprocess_h36m.py

[Ared521]

When running preprocess_h36m.py, after extracting s1-s11, I reported the following problems:

File "preprocess_h36m.py", line 141, in generate_annotations = self.load_annotations( File "preprocess_h36m.py", line 308, in load_annotations with pycdf.CDF( File "D:\Software\Anaconda\envs\mmPose\lib\site-packages\spacepy\pycdf_init.py", line 1778, in init self.open(True if readonly is None else readonly) File "D:\Software\Anaconda\envs\mmPose\lib\site-packages\spacepy\pycdf_init.py", line 1960, in open lib.call(const.OPEN, const.CDF, self.pathname, ctypes.byref(self.handle)) File "D:\Software\Anaconda\envs\mmPose\lib\site-packages\spacepy\pycdf_init.py", line 642, in call return self.check_status(self.library.CDFlib( File "D:\Software\Anaconda\envs\mmPose\lib\site-packages\spacepy\pycdf_init.py", line 595, in check_status raise CDFError(status) spacepy.pycdf.CDFError: NO_SUCH_CDF: The specified CDF does not exist.

What should I do, please? I want to generate fps10 and fps50 using metadata.xml v1.0. Because I don't have v1.2.

[ly015]

Thanks for the feedback. We will check this issue soon. Are these metadata.xml and data files obtained from the Human3.6M official site?

[Ared521]

Thanks for the feedback. We will check this issue soon. Are these metadata.xml and data files obtained from the Human3.6M official site?

Hello, thank you for your reply. I found the reason that I did not create the processed folder. Now I have a problem. After preprocessing, I train.py -- config videopose3d_ h36m_ 243frames.....py。 I only use annotation_ Body3d is in data/h36m. The images folder was not created, and the images obtained from preprocessing were not placed in the data. Why can I also run and train? Don't put pictures under data/h36m.images?

[ly015]

Videopose3D is a pose-lifting algorithm that lifts 2D keypoint coordinates to 3D without using image data. Details can be found in the paper: https://arxiv.org/abs/1811.11742

[Ared521]

Videopose3D is a pose-lifting algorithm that lifts 2D keypoint coordinates to 3D without using image data. Details can be found in the paper: https://arxiv.org/abs/1811.11742

So if I want to modify the videopose3d model and then train, all I have to do is annotations_body3d, right?