Project tracking data assets/project_graph.json is turned into a graph by
- Need to
brew install ffmpeg - https://ytcutter.link/
- Yulia Lipnistskaya: https://www.youtube.com/watch?v=ke0iusvydl8&t=337s&ab_channel=Olympics
- 1'30'' - 2'00''
-
OpenPose (by CMU) Pros: Very accurate, multi-person tracking, supports full-body and hand tracking. Cons: Requires a powerful GPU, setup can be complex. Best For: Detailed pose estimation with high accuracy.
-
MediaPipe Pose (by Google) Pros: Easy to use, works on CPU, good accuracy. Cons: Not as precise as OpenPose for detailed sports analysis. Best For: Simpler sports movements, lightweight applications.
-
DeepLabCut Pros: Highly accurate, customizable with deep learning models. Cons: Requires training a model for best results. Best For: Scientific and biomechanics research, tracking custom keypoints.
-
AlphaPose Pros: High accuracy, better handling of occlusions than OpenPose. Cons: Slower than some real-time solutions. Best For: Sports with complex poses and fast motion.
-
MoveNet (by Google) Pros: Fast, good accuracy, easy to use. Cons: Slightly lower accuracy than OpenPose/AlphaPose. Best For: Lightweight applications, fast processing.
-
BlazePose (by Google) Pros: High accuracy for human motion, optimized for mobile and desktop. Cons: May require fine-tuning for specific sports. Best For: Motion analysis in fitness, yoga, and general sports.