Research map / 3D vision / 3D generation
VoxAnchor: Explicit Voxel-Semantic Grounding for Spatial Understanding in Videos
Xinglin Li, TingTing Long, Jingzhi Zhou, Chuxuan Zeng, Jiajing Chen, Jian Yang, Jin Xie
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 Depth Anything 3 geometric features to Qwen3-VL-SFT (ablation) | Qwen3-VL-4B · VSTI-Bench (3D foundation model ablation) | average score: 54.4 -> 63.0 | Improved |
| use DA3 instead of VGGT (ablation) | VoxAnchor · VSTI-Bench (3D foundation model ablation) | average score: 59.5 -> 63.0 | Improved |
| use DA3 instead of IGGT (ablation) | VoxAnchor · VSTI-Bench (3D foundation model ablation) | average score: 60.7 -> 63.0 | Improved |
| replace standard cross-attention with Voxel-Aware Attention (ablation) | VoxAnchor · VSTI-Bench (component ablation) | average score: 63.0 -> 64.5 | Improved |
| add Geo-Semantic Adapter (ablation) | VoxAnchor · VSTI-Bench (component ablation) | average score: 64.5 -> 66.3 | Improved |
| add Voxel-Aware Attention (ablation) | VoxAnchor · VSTI-Bench (component ablation) | camera-object absolute distance: 40.0 -> 39.4 | Hurt |
| add Voxel-Aware Attention (ablation) | VoxAnchor · VSTI-Bench (component ablation) | camera movement direction: 73.2 -> 79.2 | Improved |
| add Geo-Semantic Adapter (ablation) | VoxAnchor · VSTI-Bench (component ablation) | camera-object relative distance: 73.0 -> 71.9 | Hurt |
| VoxAnchor-4B vs larger VLM-3R-7B | VoxAnchor · VSTI-Bench | average score: 58.8 -> 66.3 | Improved |
| VoxAnchor-4B vs GS-Reasoner-7B on VSI | VoxAnchor · VSI-Bench | average score: 64.7 -> 66.8 | Improved |
| VoxAnchor-4B vs GS-Reasoner-7B on absolute distance | VoxAnchor · VSI-Bench (Abs. Dist.) | MRA: 61.9 -> 52.6 | Hurt |
Closest papers by meaning
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · similarity 0.89)
- Grounded 3D-Aware Spatial Vision-Language Modeling (CVPR 2026 · similarity 0.89)
- From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D (NeurIPS 2025 · similarity 0.89)
- GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models (ICLR 2026 · similarity 0.89)
- G^3-LQ: Marrying Hyperbolic Alignment with Explicit Semantic-Geometric Modeling for 3D Visual Grounding (CVPR 2024 · similarity 0.89)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (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.
