NVIDIA has a fairly decent model on Human Keypoint extraction - BodyPose3DNet BodyPose3DNet | NVIDIA NGC to extract 34 keypoints on a human.
The model is not highly accurate hence requires additional training data. Since 34 keypoints is NOT a standard format for human keypoint extraction, I don’t have a way to train the model further.
The next option is SDG for which I was considering Omniverse, do we have a workflow or an extension which can generate synthetic humans in various environments and poses and auto generate the 34 keypoint labels?
Can this be a feature request?