Research map / Vision-language and multimodal / Multimodal LLMs
CabinSI: Omni-Cabin Spatial Reasoning through Explicit Visual Cognitive Maps
Mengxue Qu, Hengrui Hu, Mingming Ma, Ming Lei, Jie Gao, Henghui Ding, Yao Zhao, Kenn wu, Yunchao Wei
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 cognitive map framework | Qwen2.5-VL-7B · RelCabin overall | accuracy (%): 25.31 -> 41.17 | Improved |
| add cognitive map framework | Qwen3-VL-4B · RelCabin overall | accuracy (%): 27.17 -> 44.46 | Improved |
| add cognitive map framework | Qwen3-VL-8B · RelCabin overall | accuracy (%): 33.59 -> 45.92 | Improved |
| add cognitive map framework | Qwen3-VL-8B · RelCabin in-cabin multi-view | accuracy (%): 19.80 -> 24.75 | Improved |
| add cognitive map framework | Qwen2.5-VL-7B · RelCabin in-cabin multi-view | accuracy (%): 16.83 -> 26.73 | Improved |
| filter to target view only (ablation) | Rule-extracted image CogMap · RelCabin out-cabin multi-view | accuracy (%): 32.99 -> 33.15 | No clear effect |
| replace rule-based object extraction with GPT-extracted objects (ablation) | CogMap image only · RelCabin out-cabin multi-view | accuracy (%): 37.48 -> 47.46 | Improved |
| remove raw multi-view images, give cognitive map image only (ablation) | GPT-extracted CogMap · RelCabin out-cabin multi-view | accuracy (%): 35.06 -> 47.46 | Improved |
| add text form of map alongside image (ablation) | Target-view GPT-extracted CogMap · RelCabin out-cabin multi-view | accuracy (%): 35.06 -> 33.20 | Hurt |
| add proposal generation with selection (ablation) | Qwen2.5-VL-7B · RefCabin in-cabin multi-view | localization accuracy (%): 14.95 -> 22.16 | Improved |
| add cognitive map construction on top of proposals (ablation) | Qwen2.5-VL-7B · RefCabin in-cabin multi-view | localization accuracy (%): 22.16 -> 24.23 | Improved |
| add cognitive map construction on top of proposals (ablation) | Qwen2.5-VL-7B · RefCabin in-cabin single-view | localization accuracy (%): 48.26 -> 52.90 | Improved |
Closest papers by meaning
- From Indoor to Open World: Revealing the Spatial Reasoning Gap in MLLMs (CVPR 2026 · similarity 0.88)
- Seeing from Another Perspective: Evaluating Multi-View Understanding in MLLMs (AAAI 2026 · similarity 0.87)
- InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.87)
- Spatial-DISE: A Unified Benchmark for Evaluating Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.87)
- Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces (ICML 2026 · similarity 0.87)
- MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence (ICLR 2026 · similarity 0.87)
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.
