Micro-gesture recognition on iMiGUE-3K labeled
71.6Top-1 AccuracyMG-FM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MG-FMModality=Skeleton + RGB2026.05 | 71.6 | 95.07 | |
| MM-GestureModality=Skeleton + RGB2026.05 | 71.42 | — | |
| Prototype LearningModality=Skeleton + RGB2026.05 | 70.25 | — | |
| Res2Net3DModality=Skeleton + RGB2026.05 | 70.19 | 93.69 | |
| MG-FM-RGBModality=RGB-based, Evaluation Protocol=Supervised2026.05 | 69.42 | 93.08 | |
| CLIP + 3DCNNModality=Skeleton + RGB2026.05 | 68.9 | 92.43 | |
| MMGesture (RGB-only)Modality=RGB-based, Evaluation Protocol=Supervised2026.05 | 66.62 | — | |
| CLIP-MGModality=RGB-based, Evaluation Protocol=Supervised2026.05 | 61.82 | — | |
| Temporal Shift + RGB (TSM)Modality=RGB-based, Evaluation Protocol=Supervised2026.05 | 58.77 | — | |
| BlockGCN (4-stream)Modality=Skeleton-based, Stream Type=4-stream, Evaluation Protocol=Supervised2026.05 | 57.85 | 86.51 | |
| MG-FM-Skele (3-stream)Modality=Skeleton-based, Stream Type=3-stream, Evaluation Protocol=Supervised2026.05 | 57.52 | 86.78 | |
| PoseC3D (joint)Modality=Skeleton-based, Stream Type=joint, Evaluation Protocol=Supervised2026.05 | 56.13 | 85.4 | |
| Temporal Relational + RGB (TRN)Modality=RGB-based, Evaluation Protocol=Supervised2026.05 | 55.24 | — | |
| MG-FM-Skele (joint)Modality=Skeleton-based, Stream Type=joint, Evaluation Protocol=Supervised2026.05 | 53.16 | 85.45 | |
| MG-FM-Skele (bone)Modality=Skeleton-based, Stream Type=bone, Evaluation Protocol=Supervised2026.05 | 52.73 | 85.14 | |
| BlockGCN (motion)Modality=Skeleton-based, Stream Type=motion, Evaluation Protocol=Supervised2026.05 | 52.18 | 81.81 | |
| BlockGCN (bone)Modality=Skeleton-based, Stream Type=bone, Evaluation Protocol=Supervised2026.05 | 52.09 | 82.74 | |
| BlockGCN (joint)Modality=Skeleton-based, Stream Type=joint, Evaluation Protocol=Supervised2026.05 | 51.94 | 82.96 | |
| MG-FM-Skele (motion)Modality=Skeleton-based, Stream Type=motion, Evaluation Protocol=Supervised2026.05 | 51.62 | 82.76 |