Research map / 3D vision / 3D generation
AutoPhyX: Automatic Text-Condition Physics Property Generation
Bei Huang, Yixin Chen, Hongbin Zha, Yuru Pei, Siyuan Huang
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 |
|---|---|---|---|
| AutoPhyX with OpenCLIP instead of Pixie with CLIP features | Pixie · Text2PhyX test | PSNR: 23.421 -> 26.012 | Improved |
| use OpenCLIP features instead of CLIP (ablation) | AutoPhyX · Text2PhyX test | PSNR: 25.105 -> 26.012 | Improved |
| use OpenCLIP features instead of CLIP (ablation) | AutoPhyX · Text2PhyX test | Avg. physical error: 0.069 -> 0.059 | Improved |
| add text conditioning via FiLM (ablation) | AutoPhyX · Human-annotated test (30 objects) | log E error: 0.624 -> 0.181 | Improved |
| add text conditioning via FiLM (ablation) | AutoPhyX · Human-annotated test (30 objects) | log rho error: 0.287 -> 0.102 | Improved |
| use FiLM U-Net instead of 3D Transformer (16^3) (ablation) | AutoPhyX · Text2PhyX subset | log E error: 0.142 -> 0.054 | Improved |
| use FiLM U-Net instead of VisualRWKV (32^3) (ablation) | AutoPhyX · Text2PhyX subset | log E error: 0.106 -> 0.054 | Improved |
| use FiLM U-Net instead of MLP with concatenated input (ablation) | AutoPhyX · Text2PhyX subset | log E error: 0.316 -> 0.054 | Improved |
| raise VisualRWKV input resolution from 16^3 to 32^3 (ablation) | VisualRWKV · Text2PhyX subset | log E error: 0.149 -> 0.106 | Improved |
| AutoPhyX versus Gemini VLM direct prediction | Gemini · Text2PhyX test | Avg. physical error: 0.218 -> 0.059 | Improved |
| unconditional AutoPhyX versus Pixie on Pixie dataset | Pixie · Pixie dataset (unconditional) | log E error: 0.022 -> 0.063 | Hurt |
| unconditional AutoPhyX versus Pixie | Pixie · Text2PhyX (unconditional) | log E error: 0.059 -> 0.042 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VoMP — VoMP: Predicting Volumetric Mechanical Property Fields (ICLR 2026) · better on 2 of 2 shared results
- PhysX-Anything — PhysX-Anything: Simulation-Ready Physical 3D Assets from Single Image (CVPR 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PhysX-3D: Physical-Grounded 3D Asset Generation (NeurIPS 2025 · similarity 0.88)
- UniPixie: Unified and Probabilistic 3D Physics Learning via Flow Matching (CVPR 2026 · similarity 0.88)
- MotionPhysics: Learnable Motion Distillation for Text-Guided Simulation (AAAI 2026 · similarity 0.88)
- OmniPhysGS: 3D Constitutive Gaussians for General Physics-Based Dynamics Generation (ICLR 2025 · similarity 0.87)
- VoMP: Predicting Volumetric Mechanical Property Fields (ICLR 2026 · similarity 0.87)
- PhysGaia: A Physics-aware Benchmark with Multi-Body Interactions for Dynamic Novel View Synthesis (CVPR 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.
