Open AbhishekBose opened 2 years ago
The pre-trained PoseNet is for human/quadruped animals. For other categories, initial viewpoint estimation is needed (also see "sensitivity to initial root poses" in the paper supplement).
Also, this codebase uses (slow) test-time optimization and does not support feedforwd inference. (see this issue)
Can I use the pre-trained models on videos containing food items?