Research map / Vision-language and multimodal / Multimodal LLMs
VIEW2SPACE: Studying Multi-View Visual Reasoning from Sparse Observations
Fucai Ke, Zhixi Cai, Boying Li, Long Chen, Beibei Lin, Weiqing Wang, Pari Delir Haghighi, Gholamreza Haffari, Hamid Rezatofighi
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 grounded CoT with visual evidence (vs instruction tuning) (ablation) | Qwen3-VL-4B · VIEW2SPACE-v1 | MCQ accuracy (%): 36.50 -> 64.93 | Improved |
| add grounded CoT with visual evidence (vs instruction tuning) (ablation) | Qwen3-VL-4B · VIEW2SPACE-v1 | grounding mIoU (%): 47.58 -> 69.34 | Improved |
| add bounding-box grounding to CoT (vs standard CoT) (ablation) | Qwen3-VL-4B · VIEW2SPACE-v1 | grounding mIoU (%): 49.77 -> 69.34 | Improved |
| add standard CoT training (vs instruction tuning) (ablation) | Qwen3-VL-4B · VIEW2SPACE-v1 | MCQ accuracy (%): 36.50 -> 59.57 | Improved |
| add standard CoT training (vs instruction tuning) (ablation) | Qwen3-VL-4B · VIEW2SPACE-v1 | grounding mIoU (%): 47.58 -> 49.77 | Improved |
| train with CoT but answer directly at test (ablation) | Qwen3-VL-4B grounded CoT · VIEW2SPACE-v1 | MCQ accuracy (%): 64.93 -> 33.95 | Hurt |
| train on VIEW2SPACE and test on MindCube without tuning | Qwen3-VL-4B grounded CoT · MindCube-Tiny | accuracy (%): 60.76 -> 70.00 | Improved |
| enforce explicit thinking at inference (ablation) | open-source MLLMs · VIEW2SPACE-v1 | MCQ accuracy (%): Qwen3-VL-8B 31.79 -> 37.41; Qwen3-VL-4B 32.79 -> 35.19; Gemma 32.50 -> 31.14 | Mixed |
| upgrade GPT-5 to GPT-5.2 | GPT · VIEW2SPACE-v1 | MCQ accuracy (%): 59.86 -> 37.07 | Hurt |
| train on 300K data instead of smaller data / scale dataset beyond 50K (ablation) | Qwen3-VL-2B/4B · VIEW2SPACE reasoning difficulty analysis | accuracy | No clear effect |
| use open-source spatial models vs general MLLMs | Spatial models · VIEW2SPACE-v1 | MCQ accuracy (%): Spatial-MLLM 27.93 vs Qwen3-VL-8B 37.41 | No clear effect |
Closest papers by meaning
- Seeing from Another Perspective: Evaluating Multi-View Understanding in MLLMs (AAAI 2026 · similarity 0.91)
- MVVBench: Benchmarking 4D Reasoning in Vision-Language Models (NeurIPS 2026 · similarity 0.91)
- From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D (NeurIPS 2025 · similarity 0.91)
- Long Grounded Thoughts: Synthesizing Visual Problems and Reasoning Chains at Scale (ICML 2026 · similarity 0.90)
- MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods (NeurIPS 2026 · similarity 0.90)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.90)
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.
