Research map / 3D vision / 3D generation
Specular-to-Diffuse Translation for Multi-View Reconstruction
Shihao Wu, Hui Huang, Tiziano Portenier, Matan Sela, Daniel Cohen-Or, Ron Kimmel, Matthias Zwicker
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 S2Dnet translation instead of reconstructing directly from glossy images | multi-view surface reconstruction · Synthetic scenes (average) | reconstruction error (% bbox diagonal): 0.90 -> 0.64 | Improved |
| use multi-view S2Dnet instead of single-view cycleGAN | specular-to-diffuse translation · Synthetic scenes (average) | reconstruction error (% bbox diagonal): 0.80 -> 0.64 | Improved |
| use cycleGAN translation instead of reconstructing from glossy images | multi-view surface reconstruction · Synthetic scenes (average) | reconstruction error (% bbox diagonal): 0.90 -> 0.80 | Improved |
| use cycleGAN translation instead of glossy input on scene 1 | multi-view surface reconstruction · Synthetic scene 1 | reconstruction error (% bbox diagonal): 0.67 -> 1.18 | Hurt |
| use S2Dnet instead of cycleGAN for image translation | specular-to-diffuse translation · Synthetic test set | image MSE: 69.15 -> 57.78 | Improved |
| use supervised pix2pix instead of unsupervised S2Dnet | specular-to-diffuse translation · Synthetic test set | image MSE: 57.78 -> 56.20 | Improved |
| add perceptual correspondence loss (lambda_corr = 5 instead of 0) (ablation) | S2Dnet · Synthetic sequence (qualitative) | inter-view consistency | Improved |
| concatenate views along spatial axis instead of stacking pixel-wise (ablation) | S2Dnet · Synthetic data (qualitative) | translation sharpness/consistency | Improved |
| fine-tune or train on small real-world dataset | S2Dnet · Real-world objects not seen in training | translation quality | Hurt |
| use U-Net, ResNet or RNN-block generator (ablation) | S2Dnet · Synthetic data | translation quality | No clear effect |
| use paired (ConditionalGAN) instead of unpaired (cycleGAN) training (ablation) | S2Dnet · Synthetic data | translation quality | No clear effect |
Closest papers by meaning
- Transformation-Grounded Image Generation Network for Novel 3D View Synthesis (CVPR 2017 · similarity 0.88)
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.88)
- EscherNet: A Generative Model for Scalable View Synthesis (CVPR 2024 · similarity 0.87)
- Zero-Shot Novel View and Depth Synthesis with Multi-View Geometric Diffusion (CVPR 2025 · similarity 0.86)
- MVD-Fusion: Single-view 3D via Depth-consistent Multi-view Generation (CVPR 2024 · similarity 0.86)
- Understanding Multi-View Transformers (NeurIPS 2026 · 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.
