Research map / 3D vision / Face and shape
Seeing a Rose in Five Thousand Ways
Yunzhi Zhang, Shangzhe Wu, Noah Snavely, Jiajun Wu
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 |
|---|---|---|---|
| Represent object-type variation with learned distributions of 3D shape and albedo, disentangled from pose and lighting | Object-intrinsics generation from a single image · Real captured scenes, held-out instances | Depth MSE: 4.07 -> 1.10 | Improved |
| Represent object-type variation with learned distributions of 3D shape and albedo, disentangled from pose and lighting | Object-intrinsics generation from a single image · Synthetic scenes | Albedo PSNR: 18.24 -> 22.42 | Improved |
| Use the proposed model rather than NeRF for view synthesis | Object-intrinsics generation from a single image · Synthetic scenes | View synthesis PSNR: 29.95 -> 29.13 | Hurt |
| Use the proposed model rather than Neural-PIL for relighting | Object-intrinsics generation from a single image · Synthetic scenes | Relighting PSNR: 24.86 -> 25.94 | Improved |
| Apply random 2D translation and scaling augmentations to image crops | Adversarial training · Single-image instance learning | Training stability and robustness | Improved |
| Add a mask discriminator | Adversarial training · Single-image instance learning | Training stability | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Neural-PIL — Neural-PIL: Neural Pre-Integrated Lighting for Reflectance Decomposition (NeurIPS 2021) · better on 7 of 7 shared results
- GNeRF — GNeRF: GAN-Based Neural Radiance Field Without Posed Camera (ICCV 2021) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Birth and Death of a Rose (CVPR 2025 · similarity 0.89)
- Learning Non-Lambertian Object Intrinsics Across ShapeNet Categories (CVPR 2017 · similarity 0.84)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.83)
- Intrinsic Image Diffusion for Indoor Single-view Material Estimation (CVPR 2024 · similarity 0.82)
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.82)
- Latent Intrinsics Emerge from Training to Relight (NeurIPS 2024 · similarity 0.82)
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.
