Hand Pose Estimation on EgoExo4D 1.0 (test)
13.04PA-MPJPE (mm)HaMeR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HaMeRInput=Monocular2024.10 | 13.04 | 237.9 | — | |
| EgoAllo-Wrist3DInput=SLAM cameras, Losses=HaMeR + 3D wrist pose2024.10 | 14.38 | 60.08 | — | |
| EgoAllo-MonoInput=Monocular, Reprojection Guidance=Enabled2024.10 | 14.71 | 131.45 | — | |
| EgoAllo-NoReprojReprojection Guidance=None2024.10 | 14.75 | 143.2 | — | |
| ExpertAF w/ full-supprivileged_input=demo video at inference2024.08 | 131 | — | — | |
| ExpertAF2024.08 | 135 | — | 3.2 | |
| ExpertAF w/o videoablation=no video2024.08 | 139 | — | — | |
| LLaVA-FT w/ posemode=Fine-tuned, pose_input=true2024.08 | 150 | — | 1.9 | |
| ExpertAF w/ global poseablation=global pose2024.08 | 150 | — | — | |
| ExpertAF w/o alignmentablation=no temporal alignment2024.08 | 153 | — | — | |
| InternVideo2-FTmode=Fine-tuned2024.08 | 157 | — | 2 | |
| InternVideo2-NNmode=Nearest Neighbor2024.08 | 161 | — | — | |
| PoseScript/Fix2024.08 | 182 | — | 2.4 | |
| LLaVA2024.08 | — | — | 1.1 |