This week, I spent time looking into MediaPipe’s holistic landmarks detection (combining components of the pose, face, and hand landmarks to analyze full-body gestures, poses, and facial expression) to gauge if there’s a difference in computation time in comparison to their hand gesture recognition model. At the moment, what i could apply the holistic detection model to seemed to be fairly buggy on both image and video sources. In working with various MediaPipe solutions, I saw there was a scheduled update for all MediaPipe algorithms within the next week. All their recognition models have been updated with the exception of the holistic landmark detection model. the update is scheduled for March 1st, 2024 at latest so I plan to come back to this endeavor by next week’s status report.
If the update to the holistic detection model continues to fail as photographed below, I plan on trying to customize the existing models to meet the input needs of ASL.
