Questions about the model used to extract the 2D keypoints.

Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"

Apache License 2.0

1.06k stars 131 forks source link

Questions about the model used to extract the 2D keypoints. #83

Closed BellXP closed 1 year ago

BellXP commented 1 year ago

In docs/inference.md, the 2D keypoints extracted by the Fast Pose model trained on the Halpe dataset seem to have 26 keypoints, which conflict with the 17 keypoints used in MotionBERT's config file.

Is there any wrong with the model used to extract the 2D keypoints?

Walter0807 commented 1 year ago

We convert Halpe (26 keypoints) to Human 3.6M (17 keypoints).