Research map / Vision-language and multimodal / Multimodal LLMs
Physics Context Builders: A Modular Framework for Physical Reasoning in Vision-Language Models
Vahid Balazadeh, Mohammadmehdi Ataei, Hyunmin Cheong, Amir Hosein Khasahmadi, Rahul G. Krishnan
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 PCB (human narration) context | GPT-4o · Falling Tower sim, tower stable | accuracy: 59.6 -> 85.1 (+25.5) | Improved |
| add PCB (human narration) context | GPT-4o-mini · Falling Tower sim, tower stable | accuracy: 53.1 -> 84.7 (+31.6) | Improved |
| add PCB (human narration) context | Gemini 1.5 Pro · Falling Tower real, tower stable | accuracy: 60.0 -> 57.9 (-2.1) | Hurt |
| add PCB (human narration) context | GPT-4o · Falling Tower real, tower stable | accuracy: 55.0 -> 70.0 (+15.0) | Improved |
| add PCB (human narration) context | Gemini 1.5 Pro · Falling Tower sim, object above/below | accuracy: 97.8 -> 97.4 (-0.4) | No clear effect |
| add PCB (human narration) context | GPT-4o-mini · CLEVRER descriptive | accuracy: 49.5 -> 65.7 (+16.2) | Improved |
| add PCB (human narration) context | Gemini 1.5 Pro · CLEVRER explanatory per question | accuracy: 15.7 -> 35.6 (+19.9) | Improved |
| add PCB (human narration) context | GPT-4o-mini · CLEVRER counterfactual per question | accuracy: 15.6 -> 17.3 (+1.7) | Improved |
| use human narration instead of structured physics context (ablation) | GPT-4o · CLEVRER explanatory per option | accuracy: 61.1 -> 67.0 | Improved |
| use human narration instead of structured physics context (ablation) | GPT-4o · CLEVRER counterfactual per question | accuracy: 19.2 -> 28.2 | Improved |
| add descriptive QAs to stability QAs in fine-tuning (ablation) | PaliGemma-3B · Falling Tower sim, obj above/below | accuracy: 41.2 -> 100.0 | Improved |
| add stability QAs to descriptive QAs in fine-tuning (ablation) | PaliGemma-3B · Falling Tower sim, tower stable | accuracy: 39.1 -> 87.6 | Improved |
Closest papers by meaning
- Beyond Static Vision: Scene Dynamic Field Unlocks Intuitive Physics Understanding in Multi-modal Large Language Models (ICLR 2026 · similarity 0.91)
- QUANTIPHY: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models (CVPR 2026 · similarity 0.90)
- PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (ICLR 2025 · similarity 0.90)
- SeePhys: Does Seeing Help Thinking? – Benchmarking Vision-Based Physics Reasoning (NeurIPS 2025 · similarity 0.89)
- From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models (ICML 2026 · similarity 0.89)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.88)
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.
