DeepLabCut video metadata is wrongly assigned for our test datasets

[h-mayorquin]

Currently the corresponding video is determined in two steps. First, a stem is calculated here:

neuroconv/src/neuroconv/datainterfaces/behavior/deeplabcut/_dlc_utils.py

Line 182 in d1e3815

vidname, scorer = h5file.stem.split("DLC")

Then, the configuration file is used to find the videos in configuration that contain that stem:

neuroconv/src/neuroconv/datainterfaces/behavior/deeplabcut/_dlc_utils.py

Lines 211 to 214 in d1e3815

	for video_path, params in cfg["video_sets"].items():
	if vidname in video_path:
	video = video_path, params["crop"]
	break

But, if we look at the test data that we had for a while the video set paths in the configuration look like this:

video_sets:
  /Data/openfield-Pranav-2018-08-20/videos/m1s1.mp4:
    crop: 0, 640, 0, 480
  /Data/openfield-Pranav-2018-08-20/videos/m1s2.mp4:
    crop: 0, 640, 0, 480
  /Data/openfield-Pranav-2018-08-20/videos/m2s1.mp4:
    crop: 0, 640, 0, 480
  /Data/openfield-Pranav-2018-08-20/videos/m3s1.mp4:
    crop: 0, 640, 0, 480
  /Data/openfield-Pranav-2018-08-20/videos/m3s2.mp4:
    crop: 0, 640, 0, 480
  /Data/openfield-Pranav-2018-08-20/videos/m4s1.mp4:
    crop: 0, 640, 0, 480
  /Data/openfield-Pranav-2018-08-20/videos/m5s1.mp4:
    crop: 0, 800, 0, 800
  /Data/openfield-Pranav-2018-08-20/videos/m6s1.mp4:
    crop: 0, 800, 0, 800
  /Data/openfield-Pranav-2018-08-20/videos/m6s2.mp4:
    crop: 0, 800, 0, 800
  /Data/openfield-Pranav-2018-08-20/videos/m7s1.mp4:
    crop: 0, 800, 0, 800
  /Data/openfield-Pranav-2018-08-20/videos/m7s2.mp4:
    crop: 0, 800, 0, 800
  /Data/openfield-Pranav-2018-08-20/videos/m7s3.mp4:
    crop: 0, 800, 0, 800
  /Data/openfield-Pranav-2018-08-20/videos/m8s1.mp4:
    crop: 0, 800, 0, 800

And the hdf5 file name looks like this:

m3v1mp4DLC_resnet50_openfieldAug20shuffle1_30000.h5

So the stem would be m3v1mp4 but that is not to be found among that list.

I just trained some data this afternoon myself and there is the same mismatch.

https://gin.g-node.org/CatalystNeuro/behavior_testing_data/pulls/13

Maybe this is an artifact of those limited dataset but we were already thinking on passing the video_file_path as an optional argument so I think this is yet another reason.