Research map / Label-efficient and robust learning / Domain adaptation and generalization
Photo-Realistic Neural Domain Randomization
Sergey Zakharov, Rareș Ambruș, Vitor Guizilini, Wadim Kehl, Adrien Gaidon
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 |
|---|---|---|---|
| train on PNDR renderings instead of 1088 ray-traced images | CBOD · HB Scene 10 (new lighting) | ADD AUC: 30.57 -> 40.44 | Improved |
| train on PNDR renderings instead of real images | CBOD · HB Scene 10 (new lighting) | ADD AUC: 22.24 -> 40.44 | Improved |
| use PNDR instead of 1088 ray-traced images | CBOD · HB Scene 5 (in-domain) | ADD AUC: 83.62 -> 88.18 | Improved |
| increase ray-traced training images from 1088 to 4352 | CBOD · HB Scene 10 | ADD AUC: 30.57 -> 37.35 | Improved |
| use PNDR instead of ray-traced data on LM scenes | CBOD · LM Scenes 2, 8, 15 | ADD AUC: 19.71 -> 27.33 | Improved |
| add specularity and roughness randomization to albedo (ablation) | PNDR · HB5 ablation | ADD AUC: 85.43 -> 88.18 | Improved |
| use dynamic instead of fixed lighting (ablation) | PNDR · HB5 ablation | ADD AUC: 73.04 -> 88.18 | Improved |
| add indirect lighting outputs (ablation) | PNDR · HB5 ablation | ADD AUC: 87.09 -> 88.18 | No clear effect |
| add indirect lighting outputs (ablation) | PNDR · HB5 ablation | Corr (mm): 38.42 -> 34.32 | Improved |
| train monodepth2 on PNDR instead of ray-traced | Monodepth2 · HB10 | AbsRel: 0.162 -> 0.154 | Improved |
| train PackNet-SfM on PNDR instead of ray-traced | PackNet-SfM · HB5 | AbsRel: 0.11 -> 0.082 | Improved |
| use paired Pix2PixHD translation instead of unpaired CUT | CBOD · HB Scene 5 | ADD AUC: 66.49 -> 89.59 | Improved |
Closest papers by meaning
- Ray Priors Through Reprojection: Improving Neural Radiance Fields for Novel View Extrapolation (CVPR 2022 · similarity 0.88)
- HybridNeRF: Efficient Neural Rendering via Adaptive Volumetric Surfaces (CVPR 2024 · similarity 0.87)
- PhyRecon: Physically Plausible Neural Scene Reconstruction (NeurIPS 2024 · similarity 0.86)
- Real3D: The Curious Case of Neural Scene Degeneration (AAAI 2024 · similarity 0.86)
- Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models (CVPR 2025 · similarity 0.86)
- Coarse-to-Real: Generative Rendering for Populated Dynamic Scenes (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.
