Temporal-invariant Feature Extraction on RGB-stacking
1.1Displacement Error (< δ^0)VGGT-Tracking
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VGGT-TrackingBackbone=VGGT-Tracking2026.05 | 1.1 | 15.9 | 63.6 | 24.8 | |
| DINOv3-LBackbone=DINOv3-L2026.05 | 1.3 | 11.3 | 65.6 | 23.2 | |
| VGGT-Tracking + C4GBackbone=VGGT-Tracking, Feature Enhancement=C4G2026.05 | 4.9 | 28.7 | 68 | 33 | |
| DINOv3-L + C4GBackbone=DINOv3-L, Feature Enhancement=C4G2026.05 | 6.9 | 38.3 | 74.4 | 39.3 |