Research map / Vision-language and multimodal / Multimodal LLMs
ViewFusion: Structured Spatial Thinking Chains for Multi-View Reasoning
Xingjian Tao, Yiwei Wang, Yujun Cai, Yifan Song, Jing Tang
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 |
|---|---|---|---|
| train ViewFusion (SFT+RL) on Qwen3-VL-4B-Instruct | Qwen3-VL-4B-Instruct · MMSI-Bench | accuracy (%): 30.1 -> 35.4 | Improved |
| train ViewFusion on Qwen3-VL-4B-Instruct | Qwen3-VL-4B-Instruct · MindCube | accuracy (%): 37.0 -> 77.0 | Improved |
| train ViewFusion on Qwen3-VL-4B-Instruct | Qwen3-VL-4B-Instruct · ViewSpatial-Bench | accuracy (%): 42.5 -> 45.4 | Improved |
| use ViewFusion instead of a thinking-mode model | Qwen3-VL-4B-Thinking · MMSI-Bench | accuracy (%): 29.0 -> 35.4 | Improved |
| use structured two-stage output instead of free-form reasoning under RL (ablation) | ViewFusion · MMSI-Bench | accuracy (%): 33.4 -> 35.4 | Improved |
| add GRPO after SFT (ablation) | ViewFusion · MMSI-Bench | accuracy (%): 32.4 -> 35.4 | Improved |
| add format reward (ablation) | ViewFusion · MMSI-Bench | accuracy (%): 35.0 -> 35.4 | No clear effect |
| add GRPO after SFT (ablation) | ViewFusion · MindCube | accuracy (%): 68.5 -> 77.0 | Improved |
| add GRPO after SFT (ablation) | ViewFusion · ViewSpatial-Bench | accuracy (%): 45.1 -> 45.4 | No clear effect |
| use ViewFusion | Qwen3-VL-4B-Instruct · MMSI-Bench Obj.-Reg. | accuracy (%): 29.4 -> 21.2 | Hurt |
| use ViewFusion | Qwen3-VL-4B-Instruct · MMSI-Bench Cam.-Cam. | accuracy (%): 30.1 -> 46.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SpatialLadder-3B — SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models (ICLR 2026) · better on 1 of 1 shared result
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
- From Correspondence to Actions: Human-Like Multi-Image Spatial Reasoning in Multi-modal Large Language Models (ICML 2026 · similarity 0.90)
- 3ViewSense: Spatial and Mental Perspective Reasoning from Orthographic Views in Vision-Language Models (ICML 2026 · similarity 0.90)
- Actial: Activate Spatial Reasoning Ability of Multimodal Large Language Models (NeurIPS 2025 · similarity 0.89)
- STAR-R1: Multi-View Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMs (CVPR 2026 · similarity 0.89)
- Learning Multi-View Spatial Reasoning from Cross-View Relations (CVPR 2026 · similarity 0.89)
- Multi-view Relational Distillation for Spatial Reasoning with Vision-Language Models (NeurIPS 2026 · 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.
