Research map / Vision-language and multimodal / Human and embodied modeling
Physion: Evaluating Physical Prediction from Vision in Humans and Machines
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use particle-based GNN dynamics models with ground-truth physical state instead of vision-only models | Physion OCP task · Collide | OCP accuracy: 0.597 (SVG) -> 0.85 (GNS) | Helped |
| learn object-centric contrastive latents (CSWM) instead of convolutional latents (SVG) (ablation) | unsupervised vision dynamics models · Physion all-scenario | OCP accuracy | Helped |
| add ImageNet-pretrained encoder (pVGG, pDeIT) rather than training encoder only on Physion (ablation) | vision dynamics models · Physion all-scenario | OCP accuracy | Helped |
| add simulated future features to observed features in readout (ablation) | vision models · Physion all-scenario | OCP accuracy | No clear effect |
| use full-movie readout instead of observed-only (ablation) | vision models · Physion all-scenario | OCP accuracy: b=0.094 | Helped |
| train on only one scenario instead of all scenarios (ablation) | vision models · Physion all-scenario | OCP accuracy: b=0.21 | No clear effect |
| use supervised object-centric RPIN instead of unsupervised CSWM (ablation) | vision dynamics models · Physion | OCP accuracy: b=0.035 | Mixed |
Closest papers by meaning
- Physion++: Evaluating Physical Scene Understanding that Requires Online Inference of Different Physical Properties (NeurIPS 2023 · similarity 0.93)
- PhysInOne: Visual Physics Learning and Reasoning in One Suite (CVPR 2026 · similarity 0.90)
- PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (ICLR 2025 · similarity 0.89)
- PhysPT: Physics-aware Pretrained Transformer for Estimating Human Dynamics from Monocular Videos (CVPR 2024 · similarity 0.88)
- PhysX-3D: Physical-Grounded 3D Asset Generation (NeurIPS 2025 · similarity 0.87)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · 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.