Closed frankier closed 3 years ago
We only need to embed a few of the best chips within a particular shot to identify who they are. We should pick the best frames and use those for identification. We should do this so we still do efficient batching on the GPU.
This only makes sense for OpenPose.
This snippet has some nice ideas for face keyframe selection. is_blurred might be a good idea:
https://gist.github.com/facepainter/adfaabe25831a7c9300bafd1b886e1c8
We only need to embed a few of the best chips within a particular shot to identify who they are. We should pick the best frames and use those for identification. We should do this so we still do efficient batching on the GPU.
This only makes sense for OpenPose.