src/datasets/handcrop_poseflow.py duplicates path

[RodGal-2020]

Launching the training code I received the following error:

At first I believed that I could have forgotten to move the keypoints, but a closer inspection revealed the duplicaiton of part of the path. Following this, adding some "print" lines in the src/datasets/handcrop_poseflow.py file I got:

As you can see in the image, sample['path'].replace('mp4', 'kp') already includes the data from previous objects, i.e., self.root_path.replace('mp4', 'kp') and self.join_path.

I believe that this is a typo, as with the following code in line ~107 in src/datasets/handcrop_poseflow.py the kp_path object works fine:

           kp_path = os.path.join(
                                   #self.root_path.replace('mp4', 'kp'), 
                                   #self.job_path,
                                   sample['path'].replace('mp4', 'kp'), 
                                   '{}_{:012d}_keypoints.json'.format(sample['path'].split('/')[-1].replace('.mp4', ''), frame_index))

[RodGal-2020]

I believe that this also happens with the line ~104, where you can find:

frames, _, _ = torchvision.io.read_video(os.path.join(self.root_path, self.job_path, sample['path']),
                                                 pts_unit='sec')

In which case sample['path'] already includes the previous objects.

Example: