search

IanMagnusson / Wav2Lip-Emotion

Wav2Lip-Emotion extends Wav2Lip to modify facial expressions of emotions via L1 reconstruction and pre-trained emotion objectives. We also propose a novel automatic evaluation for emotion modification corroborated with a user study.
94 stars 17 forks source link

make real_videos_inference.py preserve video names #1

Closed IanMagnusson closed 3 years ago

IanMagnusson commented 3 years ago

Right now real_videos_inference.py names its output files solely as numbers based on the order of files in the file list. This is a pain because changing the file list can clobber outputs, and it you need the file lists to backtrack a generated video to its origin.

Instead it should use the original file name and maybe add some extra unique identifier.

The offending line of code is here: https://github.com/jagnusson/emotion-synthesis/blob/4ebbabb68bece84397cd0bb761f3686b645db2d3/evaluation/real_videos_inference.py#L298

Ian will handle this when he has a chance.

IanMagnusson commented 3 years ago

Both gen_videos_from_filelist.py and real_videos_inference.py use names of input audio and video to name the output file. Note this will be redundant in the case of generated outputs on mead at inference time.