Feature/recursive videos to poses

[cleong110]

Implemented a recursive option, solves #125 . Also added in an optional argument to specify the video extension to search for, though currently the only choice is .mp4. All behavior should be the same as before, as in if you don't use the -r argument it will just look in the current directory.

[cleong110]

Did a quick edit to the file to _SUPPORTED_VIDEO_FORMATS to see how easy it would be to switch extensions, and it successfully found and ran on .webm files as expected. But .webm support is a separate issue. #127

[cleong110]

Updated code to check against list of pose files as requested, and to iterate over all video suffixes. Realized this could result in foo.mp4, foo.webm both being posed to foo.pose, so added an option to keep original video suffix, and added a "this will cause name collisions do you want to keep going" check

[cleong110]

Also added a try/except checking for ValueError, so that if one of the video files is corrupted, the utility will keep going.

Also added a check for existing .pose files, in which case it will not re-run/overwrite.

[AmitMY]

much better. still minor changes

[AmitMY]

LGTM - just revert one change and i'll merge.
Thanks!

[J22Melody]

It is great!

I will add a multiprocessing option to the script to use multiple CPUs. Currently, I need to run pose estimation on 1000+ hours of BSL data, and do not want to wait for too long ;)

[cleong110]

@J22Melody I actually did that with GNU parallel lol

[cleong110]

https://github.com/cleong110/ud-vlab/blob/main/data_munging/split_folder.py
https://github.com/cleong110/ud-vlab/blob/main/data_munging/run_videos_to_poses_on_multiple_dirs.sh

[J22Melody]

@cleong110 please share the repo with me!

[cleong110]

Oh silly me! The two files in question:
run_videos_to_poses_on_multiple_dirs.zip

[cleong110]

Not that we shouldn't add the multiprocessing! I think that's a good thing to do. This was just my workaround:

1. Split folder of many videos, into many folders of, like, 100 videos.
2. find "path/to/directory_of_directories" -mindepth 1 -maxdepth 1 -type d|parallel -j "$job_count" --progress --results "$results_prefix" videos_to_poses --format mediapipe --directory "{}"
   So then if job_count is set to 100, you can spin up 100 instances of videos_to_poses in parallel