Add video_human_3d_pose_mapper.

[Qirui-jiao]

Add video_human_3d_pose_mapper: Extract 3D human pose with Human3R (SMPL-X).

[gemini-code-assist]

Code Review

This pull request introduces the VideoHuman3DPoseMapper, an operator designed to extract 3D human poses from videos using the Human3R (SMPL-X) model. The implementation includes the core mapper logic, configuration updates, model preparation utilities, and unit tests. Review feedback identifies several improvement opportunities: correcting the default model path and YAML null values in the configuration, adopting loguru for standardized logging, fixing a typo in an error message, and addressing unreachable code and potential resource leaks in video parsing. Additionally, suggestions were made to refine the model setup process to be less disruptive to automated pipelines and to use the huggingface_hub library for more robust model downloads.

[Qirui-jiao]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces the VideoHuman3DPoseMapper for extracting 3D human poses using the Human3R (SMPL-X) model, along with model preparation utilities and documentation updates. The review feedback identifies several critical issues, including an uninitialized video_key attribute that causes runtime errors, the need to convert metadata to JSON-serializable lists, and potential race conditions in the model loading logic. Additionally, the reviewer recommended replacing assert statements with robust error handling and extending the mapper to process all videos within a sample instead of only the first one.