Research map / 3D vision / 3D generation
Unsupervised Discovery of 3D Physical Objects from Video
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 |
|---|---|---|---|
| add physical consistency (ablation) | POD-Net with multi-scale inference · ADEPT / Moving ShapeNet | IoU: 0.649 (0.011) -> 0.739 (0.011) | Helped |
| add physical consistency (ablation) | POD-Net with multi-scale inference · ADEPT / Moving ShapeNet | Detection: 0.709 (0.016) -> 0.821 (0.015) | Helped |
| add multi-scale inference (ablation) | POD-Net without physics · ADEPT / Moving ShapeNet | IoU: 0.314 (0.010) -> 0.649 (0.011) | Helped |
| add physics (ablation) | POD-Net with single-scale inference · ADEPT / Moving ShapeNet | IoU: 0.314 (0.010) -> 0.462 (0.007) | Helped |
| replace learned backprojection with manual backprojection (ablation) | POD-Net with multi-scale inference and physics · ADEPT / Moving ShapeNet | IoU: 0.739 (0.011) -> 0.685 (0.017) | Hurt |
| use physical consistency (ablation) | POD-Net · novel objects and novel colors | average IoU | Helped |
| add multi-scale inference (ablation) | POD-Net without physics · real block towers | object separation | Helped |
| add physical consistency (ablation) | POD-Net with multi-scale inference · real block towers | object separation | Helped |
| enforce physical consistency during early training rather than after object-like masks emerge (ablation) | POD-Net · training | performance | Hurt |
| use inferred physical objects for plausibility judgments | POD-Net · Block task | relative accuracy: POD-Net: 0.622; GAN: 0.44; Encoder-Decoder: 0.52; LSTM: 0.44 | Helped |
| use inferred physical objects for plausibility judgments | POD-Net · Overturn (Long) task | relative accuracy: 0.77; [cite]: 0.73; GAN: 0.81; Encoder-Decoder: 0.61; LSTM: 0.63 | Mixed |
Closest papers by meaning
- 4D Unsupervised Object Discovery (NeurIPS 2022 · similarity 0.87)
- Discovering Objects That Can Move (CVPR 2022 · similarity 0.87)
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion (CVPR 2025 · similarity 0.85)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.85)
- UNION: Unsupervised 3D Object Detection using Object Appearance-based Pseudo-Classes (NeurIPS 2024 · similarity 0.85)
- Physical Property Understanding from Language-Embedded Feature Fields (CVPR 2024 · similarity 0.84)
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.