Research map / 3D vision / 3D generation
G$^2$TAM: Geometry Grounded Track Anything Model
Chenming Zhu, Peizhou Cao, Jingli Lin, Wenbo Hu, Yunlong Ran, Jiangmiao Pang, Tai Wang, Xihui 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 early-fusion cross-modal spatial encoder instead of decoder late fusion (ablation) | G2TAM · InsTrack-Text | S-mIoU: 61.8 -> 72.3 | Improved |
| inject text tokens into all frames instead of 1 frame (ablation) | G2TAM · InsTrack-Text | S-mIoU: 58.8 -> 72.3 | Improved |
| inject text tokens into 5 frames instead of 2 (ablation) | G2TAM · InsTrack-Text | S-mIoU: 63.2 -> 67.9 | Improved |
| inject text tokens into all frames on Ref-YouTube-VOS (ablation) | G2TAM · Ref-YouTube-VOS | J&F: 59.9 -> 70.2 | Improved |
| use Pi3 geometry-pretrained encoder instead of SAM2 encoder (ablation) | G2TAM · InsTrack-Text | S-mIoU: 54.2 -> 61.8 | Improved |
| add joint reconstruction training to segmentation training (ablation) | G2TAM · InsTrack-Text | S-mIoU: 68.2 -> 72.3 | Improved |
| add joint reconstruction training to segmentation training (ablation) | G2TAM · InsTrack-Visual | S-mIoU: 72.7 -> 75.8 | Improved |
| add SAM2-style explicit memory bank (ablation) | G2TAM · InsTrack-Visual | S-mIoU: 75.8 -> 74.9 | Hurt |
| add SAM2-style explicit memory bank (ablation) | G2TAM · MOSE | J&F: 77.8 -> 78.2 | Mixed |
| add SAM2-style explicit memory bank (ablation) | G2TAM · DAVIS 2017 | J&F: 89.9 -> 89.5 | Hurt |
| shuffle temporal order of frames (ablation) | G2TAM · InsTrack-Text | S-mIoU: 72.3 -> 70.1 | Hurt |
| shuffle temporal order of frames (ablation) | G2TAM · InsTrack-Visual | S-mIoU: 75.8 -> 75.4 | No clear effect |
Closest papers by meaning
- IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer (NeurIPS 2026 · similarity 0.87)
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024 · similarity 0.86)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.86)
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · similarity 0.86)
- G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning (CVPR 2026 · similarity 0.86)
- Instance Tracking in 3D Scenes from Egocentric Videos (CVPR 2024 · similarity 0.85)
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.
