Research map / 3D vision / Depth and stereo
SpatialTrackerV2: Advancing 3D Point Tracking with Explicit Camera Motion
Yuxi Xiao, Jianyuan Wang, Nan Xue, Nikita Karaev, Yuri Makarov, Bingyi Kang, Xing Zhu, Hujun Bao, Yujun Shen, Xiaowei Zhou
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 camera pose estimation (offl+ vs offl-) (ablation) | SpatialTrackerV2 + MegaSAM · TAPVid-3D Aria subset | AJ (3D): 22.3 -> 24.7 | Improved |
| add camera pose estimation (offl+ vs offl-) (ablation) | SpatialTrackerV2 + MegaSAM · TAPVid-3D Aria subset | APD3D: 32.2 -> 35.2 | Improved |
| add camera pose estimation (offl+ vs offl-) (ablation) | SpatialTrackerV2 + MegaSAM · TAPVid-3D average | AJ (3D): 18.7 -> 19.8 | Improved |
| use MegaSAM instead of UniDepthV2 depth | CoTracker3 · TAPVid-3D average | AJ (3D): 12.8 -> 17.3 | Improved |
| replace SyncFormer with simple 3D projection layer (ablation) | CoTracker3 · TAP-Vid DAVIS | AJ: 64.4 -> 51.6 | Hurt |
| replace SyncFormer with simple 3D projection layer (ablation) | CoTracker3 · TAP-Vid DAVIS | OA: 91.2 -> 85.3 | Hurt |
| add VKITTI and PointOdyssey with joint training (ablation) | Base tracker · TAPVid-3D DriveTrack | AJ (3D): 7.4 -> 14.7 | Improved |
| add VKITTI and PointOdyssey with joint training (ablation) | Base tracker · TAPVid-3D DriveTrack | APD3D: 13.3 -> 21.9 | Improved |
| add VKITTI and PointOdyssey with joint training (ablation) | Base tracker · TAPVid-3D Aria subset | AJ (3D): 16.0 -> 15.7 | No clear effect |
| add VKITTI and PointOdyssey with joint training (ablation) | Base tracker · TAPVid-3D PStudio | AJ (3D): 12.6 -> 17.2 | Improved |
| add back-end joint motion optimization to front-end (ablation) | SpatialTrackerV2 · Camera pose Sintel | ATE: 0.075 -> 0.054 | Improved |
| add back-end joint motion optimization to front-end (ablation) | SpatialTrackerV2 · Camera pose TUM-dynamic | ATE: 0.038 -> 0.012 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- MegaSAM — MegaSaM: Accurate, Fast and Robust Structure and Motion from Casual Dynamic Videos (CVPR 2025) · better on 3 of 5 shared results
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 2 of 2 shared results
- TAPIP3D — TAPIP3D: Tracking Any Point in Persistent 3D Geometry (NeurIPS 2025) · better on 2 of 2 shared results
- SpatialTracker — SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024 · similarity 0.93)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.91)
- TrackingWorld: World-centric Monocular 3D Tracking of Almost All Pixels (NeurIPS 2025 · similarity 0.91)
- TAPIP3D: Tracking Any Point in Persistent 3D Geometry (NeurIPS 2025 · similarity 0.91)
- TAPVid-3D: A Benchmark for Tracking Any Point in 3D (NeurIPS 2024 · similarity 0.90)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.89)
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.
