feat: add MediaPipe face keypoint support #7
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This change add feature for adding face landmarks using mediapipe, the change matched the previous implementation for body and hands so the output can be skeleton pose compatible with openpose output to feed into the controlnet.
There was some performance degradation as reported by @ryanontheinside where the hands and pose detection was getting > 20 fps but with the addition of face landmarks the peformance is about 7-8 fps. We can do further testing to verify if the performance regression is the cause of main process of face landsmarks or the way the files are handled.
Also there is another route that can be followed where instead of monolith face, hand and face tracking, individual modules for each can be utilised such that if needed one can be activated and deactivated which might speedup the process according to the need.
This is a solution to the https://linear.app/livepeer/issue/ENG-2864/optimize ticket.