Research map / 3D vision / 3D generation
Predict-Optimize-Distill: A Self-Improving Cycle for 4D Object Understanding
Mingxuan Wu, Huang Huang, Justin Kerr, Chung Min Kim, Anthony Zhang, Brent Yi, Angjoo Kanazawa
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 predict-optimize-distill cycle instead of pure optimization (RSRD) | POD · Synthetic (5 objects) | PCP alpha=0.05: 0.454 -> 0.778 | Improved |
| same, mean squared error | POD · Synthetic (5 objects) | MSE: 0.0952 -> 0.0422 | Improved |
| add synthetic novel-view augmentation for distillation (ablation) | POD · Synthetic (5 objects) | PCP alpha=0.03: 0.561 -> 0.622 | Improved |
| add quasi-multiview supervision (ablation) | POD · Synthetic (5 objects) | PCP alpha=0.03: 0.570 -> 0.622 | Improved |
| add RSRD initialization of the predictor (ablation) | POD · Synthetic (5 objects) | PCP alpha=0.03: 0.603 -> 0.622 | Improved |
| add RSRD initialization of the predictor (ablation) | POD · Synthetic (5 objects) | MSE: 0.0434 -> 0.0422 | Improved |
| use longer input video (1s to 6s) | POD · Synthetic (varied video length) | PCP: performance delta of 6% | Improved |
| run more cycle iterations on longest video | POD · Synthetic (6 s video) | PCP: PCP improved by 14% | Improved |
| run more cycle iterations on short video | POD · Synthetic (short video) | PCP | No clear effect |
| longer video at loop 0 (before any cycle) | POD · Synthetic, loop 0 | PCP | Hurt |
Closest papers by meaning
- DistillNeRF: Perceiving 3D Scenes from Single-Glance Images by Distilling Neural Fields and Foundation Model Features (NeurIPS 2024 · similarity 0.87)
- Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos (CVPR 2025 · similarity 0.86)
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion (CVPR 2025 · similarity 0.86)
- Articulation in Prime: Primitive-Based Articulated Object Understanding from a Single Casual Video (NeurIPS 2026 · similarity 0.86)
- Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding (NeurIPS 2025 · similarity 0.86)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · 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.
