Research map / Detection and segmentation / Tracking
Fast Spatial Tracking with Visual Geometry Transformer
Chengjie Huang, Guile Wu, Dongfeng Bai, Bingbing Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add global embedding branch to the frame-only tracker (ablation) | Fast Spatial Tracker · TAPVid-3D minival | ADP: 14.5 -> 16.0 | Improved |
| add multi-layer cross attention (MLCA) to the frame-only tracker (ablation) | Fast Spatial Tracker · TAPVid-3D minival | ADP: 14.5 -> 15.0 | Improved |
| combine global branch and MLCA (ablation) | Fast Spatial Tracker · TAPVid-3D minival | ADP: 14.5 -> 16.3 | Improved |
| add four real-world 3D datasets (ScanNet, ARKitScenes, nuScenes, Argoverse2) to synthetic training (ablation) | Fast Spatial Tracker · TAPVid-3D minival | ADP: ADP 16.3 -> 21.9; OA 81.5 -> 76.8 | Mixed |
| fine-tune DINOv2 and alternating attention layers of the backbone with reduced learning rate (ablation) | Fast Spatial Tracker · TAPVid-3D minival | ADP: 21.9 -> 25.7 | Improved |
| add DynPose100K++ and Wild-SDG1M pseudo-labeled in-the-wild and AI-generated data (ablation) | Fast Spatial Tracker · TAPVid-3D minival | ADP: 25.7 -> 27.5 | Improved |
| switch depth prior from UniDepthV2 to MegaSaM | DELTA · TAPVid-3D | ADP: 20.6 -> 26.3 | Improved |
| switch depth prior from UniDepthV2 to MegaSaM | SpaTracker · TAPVid-3D | ADP: 16.8 -> 20.8 | Improved |
| remove depth prior and use geometry transformer features directly | Fast Spatial Tracker vs SpaTrackerV2 · TAPVid-3D inference | time per frame: 309ms -> 28ms | Improved |
| use the 3D tracker for 2D point tracking | Fast Spatial Tracker · TAP-Vid | AJ: 61.4 vs AllTracker 67.8 | Hurt |
Closest papers by meaning
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.90)
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024 · similarity 0.90)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.89)
- TrackingWorld: World-centric Monocular 3D Tracking of Almost All Pixels (NeurIPS 2025 · similarity 0.88)
- TAPIP3D: Tracking Any Point in Persistent 3D Geometry (NeurIPS 2025 · similarity 0.87)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
