Research map / Vision-language and multimodal / Multimodal LLMs
Towards Spatial Supersensing in the Wild
Tianjun Gu, Tianyu Xin, Kuan Zhang, Bowen Yang, Yinan Han, Peize Li, Yucheng Lu, Jianhang Liu, Xinran Zhang, KOK CHUNG CHUA, Qiyue Zhao, Qinlei Xie and 3 more
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 |
|---|---|---|---|
| rotate query frame viewpoint in VPO | GPT-5.4 · VSI-Super-Wild VPO | accuracy (%): 26.0% -> 24.9% | Hurt |
| rotate query frame viewpoint in VPO | Qwen3-VL-8B · VSI-Super-Wild VPO | accuracy (%): 25.7% -> 24.4% | Hurt |
| rotate query frame viewpoint in VPO | Cambrian-S-7B · VSI-Super-Wild VPO | accuracy (%): 26.2% -> 25.6% | Hurt |
| ask last-occurrence instead of first-occurrence ordering in VOO | GPT-5.4 · VSI-Super-Wild VOO | accuracy (%): 31.3% -> 22.1% | Hurt |
| ask last-occurrence instead of first-occurrence ordering in VOO | Qwen3-VL-8B · VSI-Super-Wild VOO | accuracy (%): 50.2% -> 30.5% | Hurt |
| ask last-occurrence instead of first-occurrence ordering in VOO | Cambrian-S-7B · VSI-Super-Wild VOO | accuracy (%): 50.7% -> 30.0% | Hurt |
| lengthen video duration from 0-10 to 120+ minutes | open-source models average · VSI-Super-Wild | overall score: 35.0 -> 26.3 | Hurt |
| probe object state (VOO) instead of agent (VMR) or place (VPO) state | Cambrian-S-7B · VSI-Super-Wild | accuracy (%): 26.91 -> 40.36 | Improved |
| use streaming input instead of 128 sampled frames | Cambrian-S-7B-LFP · VSI-Super-Wild | overall score: 32.86 -> 33.35 | No clear effect |
| scale Cambrian-S from 0.5B to 3B | Cambrian-S · VSI-Super-Wild | overall score: 21.46 -> 33.18 | Improved |
| scale Cambrian-S from 3B to 7B | Cambrian-S · VSI-Super-Wild | overall score: 33.18 -> 34.22 | No clear effect |
Closest papers by meaning
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.88)
- MMWorld: Towards Multi-discipline Multi-faceted World Model Evaluation in Videos (ICLR 2025 · similarity 0.87)
- Thinking in Dynamics: How Multimodal Large Language Models Perceive, Track, and Reason Dynamics in Physical 4D World (CVPR 2026 · similarity 0.87)
- OST-Bench: Evaluating the Capabilities of MLLMs in Online Spatio-temporal Scene Understanding (NeurIPS 2025 · similarity 0.86)
- DynamicVerse: A Physically-Aware Multimodal Framework for 4D World Modeling (NeurIPS 2025 · similarity 0.86)
- ERGeoBench: A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models (ICML 2026 · similarity 0.86)
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.
