The 3DPW dataset includes various complex scenarios with multiple individuals present in the frames. Since AlphaPose supports multi-person tracking, the resulting JSON file contains several annotations for each frame. How did you filter these annotations to focus only on the keypoints of the individuals (maximum two people) of interest when using MotionBERT? Are you doing "3D Pose Estimation with known Association", by using the 2D-keypoint ground truth to determine the ID of the tracked persons as described here: https://competitions.codalab.org/competitions/24938?
Thank you!
The 3DPW dataset includes various complex scenarios with multiple individuals present in the frames. Since AlphaPose supports multi-person tracking, the resulting JSON file contains several annotations for each frame. How did you filter these annotations to focus only on the keypoints of the individuals (maximum two people) of interest when using MotionBERT? Are you doing "3D Pose Estimation with known Association", by using the 2D-keypoint ground truth to determine the ID of the tracked persons as described here: https://competitions.codalab.org/competitions/24938?
Thank you!