Research map / 3D vision / Depth and stereo
Structure from Duplicates: Neural Inverse Graphics from a Pile of Objects
Tianhang Cheng, Wei-Chiu Ma, Kaiyu Guan, Antonio Torralba, Shenlong Wang
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 |
|---|---|---|---|
| use duplicate-object sharing instead of InvRender (multi-view) | inverse rendering · Multi-view synthetic | albedo PSNR: 16.984 -> 21.961 | Improved |
| use duplicate-object sharing instead of InvRender (multi-view) | inverse rendering · Multi-view synthetic | geometry Chamfer distance: 0.024 -> 0.011 | Improved |
| train on single view versus baselines on 10 views | inverse rendering · Single-view synthetic | albedo PSNR: 14.977 -> 17.629 | Improved |
| train on single view versus baselines on 10 views | inverse rendering · Single-view synthetic | geometry Chamfer distance: 0.033 -> 0.034 | No clear effect |
| single view versus multi-view baselines on real data | inverse rendering · Real-world single-view | rendering PSNR: 20.665 -> 20.326 | No clear effect |
| single view versus multi-view baselines on real data | inverse rendering · Real-world single-view | LPIPS: 0.262 -> 0.192 | Improved |
| remove mask loss (ablation) | full model · Loss ablation | albedo PSNR: 17.27 -> 15.14 | Hurt |
| remove pretrained surface normal prior (ablation) | full model · Loss ablation | albedo PSNR: 17.27 -> 16.51 | Hurt |
| remove Eikonal loss (ablation) | full model · Loss ablation | relighting PSNR: 19.95 -> 19.31 | Hurt |
| remove metallic loss (ablation) | full model · Loss ablation | metallic MSE: 0.02 -> 0.08 | Hurt |
| remove latent smooth regularization (ablation) | full model · Loss ablation | relighting PSNR: 19.95 -> 19.71 | Hurt |
| increase number of duplicates from 6 to 30 | full model · color box dataset | albedo PSNR: 18.48 -> 19.69 | Improved |
Closest papers by meaning
- ReLi3D: Relightable Multi-view 3D Reconstruction with Disentangled Illumination (ICLR 2026 · similarity 0.87)
- Share with Thy Neighbors: Single-View Reconstruction by Cross-Instance Consistency (ECCV 2022 · similarity 0.86)
- Structure from Collision (CVPR 2025 · similarity 0.86)
- Slice3D: Multi-Slice Occlusion-Revealing Single View 3D Reconstruction (CVPR 2024 · similarity 0.86)
- LIRM: Large Inverse Rendering Model for Progressive Reconstruction of Shape, Materials and View-dependent Radiance Fields (CVPR 2025 · similarity 0.85)
- The More You See in 2D the More You Perceive in 3D (CVPR 2024 · similarity 0.85)
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.
