Research map / 3D vision / Face and shape
Neural Reflectance for Shape Recovery With Shadow Handling
Junxuan Li, Hongdong Li
Check 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 neural specular reflectance modeling and explicit shadow handling in the self-supervised method | Ours · DiLiGenT (10 objects, Avg.; no ground-truth normals) | MAE: 6.50 vs 8.83 | Improved |
| Use estimated neural svBRDFs for image re-rendering | Ours · DiLiGenT (10 objects, Avg.) | PSNR: 42.93 -> 43.42 | Improved |
| Add the early-stage smoothness constraint (ablation) | Ours · DiLiGenT (10 objects, Avg.) | MAE: 7.61 -> 6.50 | Improved |
| Add explicit shadow rendering and the shadow factor (ablation) | Ours · DiLiGenT (10 objects, Avg.) | MAE: 8.44 -> 6.50 | Improved |
| Add specular reflectance modeling (ablation) | Ours · DiLiGenT (10 objects, Avg.) | MAE: 13.30 -> 6.50 | Improved |
| Output surface normals directly rather than outputting depth from the surface network (ablation) | Ours surface modeling · DiLiGenT (10 objects, Avg.) | MAE: 7.56 -> 6.50 | Improved |
| Reduce the input image count from 96 to 16 | Ours · DiLiGenT (10 objects, 16 input images) | MAE: 6.50 -> 6.82 | Hurt |
| Reduce the input image count from 96 to 10 | Ours · DiLiGenT (10 objects, 10 input images) | MAE: 6.50 -> 7.47 | Hurt |
| Reduce the input image count from 96 to 8 | Ours · DiLiGenT (10 objects, 8 input images) | MAE: 6.50 -> 7.70 | Hurt |
Closest papers by meaning
- Shape and Material Capture at Home (CVPR 2021 · similarity 0.90)
- Reflectance and Natural Illumination from Single-Material Specular Objects Using Deep Learning (TPAMI 2017 · similarity 0.89)
- Neural Inverse Rendering for General Reflectance Photometric Stereo (ICML 2018 · similarity 0.89)
- NeRSP: Neural 3D Reconstruction for Reflective Objects with Sparse Polarized Images (CVPR 2024 · similarity 0.87)
- GNeRP: Gaussian-guided Neural Reconstruction of Reflective Objects with Noisy Polarization Priors (ICLR 2024 · similarity 0.87)
- UniSDF: Unifying Neural Representations for High-Fidelity 3D Reconstruction of Complex Scenes with Reflections (NeurIPS 2024 · similarity 0.87)
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.
