[Feature] 3D pose model with keypoint scores

[mitch-spec]

What is the feature?

Examples inferencer_demo.py (pose3d=human3d) and body3d_pose_lifter_demo.py produce predictions in a JSON file with all keypoint_scores with the value 1.

"keypoint_scores": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]

Meanwhile inferencer_demo.py (pose2d=human) produces valid keypoint_scores.

"keypoint_scores": [ 0.44526207447052, 0.26142141222953796, 0.5664293766021729, 0.6183218955993652, 0.7224370241165161, 0.6289105415344238, 0.6901071071624756, 0.4989352226257324, 0.6010563373565674, 0.2657913565635681, 0.46015164256095886, 0.7015870213508606, 0.6015724539756775, 0.5971701145172119, 0.6327921748161316, 0.6255173683166504, 0.664059042930603 ],

inferencer_demo.py (pose3d=human3d) is using motionbert_dstformer-243frm_8xb32-240e_h36m.
The body3d_pose_lifter_demo.py example uses rtmpose-m_simcc-body7_pt-body7_420e-256x192 (2D) and videopose_h36m_243frames_fullconv_supervised_cpn (3D).

Is there a 3D pose model in the model zoo that provides valid keypoint_scores in 3D?

Any other context?

No response