Research map / 3D vision / 3D generation
SuperMat: Physically Consistent PBR Material Estimation at Interactive Rates
Yijia Hong, Yuan-Chen Guo, Ran Yi, Yulong Chen, Yan-Pei Cao, Lizhuang Ma
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 |
|---|---|---|---|
| add end-to-end training (trailing scheduler fix plus perceptual loss) (ablation) | SuperMat · Image-space decomposition (192 objects), albedo | PSNR: 24.26 -> 26.70 | Improved |
| add end-to-end training (trailing scheduler fix plus perceptual loss) (ablation) | SuperMat · Image-space decomposition (192 objects), inference time | time (s): 3.09 -> 0.07 | Improved |
| add re-render loss (ablation) | SuperMat · Image-space decomposition (192 objects), albedo | PSNR: 26.70 -> 27.68 | Improved |
| add re-render loss (ablation) | SuperMat · Image-space decomposition (192 objects), relighting | PSNR: 26.41 -> 27.66 | Improved |
| add re-render loss (ablation) | SuperMat · Image-space decomposition (192 objects), metallic | PSNR: 24.54 -> 25.48 | Improved |
| use perceptual loss instead of SSI loss (ablation) | SuperMat · Loss function comparison (Objaverse test, no re-render loss), albedo | PSNR: 24.2359 -> 27.0082 | Improved |
| use perceptual loss instead of L1 loss (ablation) | SuperMat · Loss function comparison (Objaverse test, no re-render loss), albedo | PSNR: 26.8102 -> 27.0082 | Improved |
| train on synthetic plus real-captured data instead of real-captured only (ablation) | SuperMat · Training data ablation (64 objects, 6144 inputs), albedo | PSNR: 24.50 -> 27.58 | Improved |
| train on synthetic plus real-captured data instead of synthetic only (ablation) | SuperMat · Training data ablation (64 objects, 6144 inputs), albedo | PSNR: 27.41 -> 27.58 | Improved |
| train on synthetic plus real-captured data instead of synthetic only (ablation) | SuperMat · Image-space decomposition (192 objects), metallic | PSNR: 23.80 -> 23.77 | No clear effect |
| add UV refinement network to blended SuperMatMV UV maps (ablation) | SuperMatMV pipeline · UV maps from SuperMatMV, albedo | PSNR: 13.56 -> 23.92 | Improved |
| replace StableMaterialMV with SuperMat pipeline for 3D relighting | UV backprojection pipeline · Novel-view relighting, BlenderVault (14 objects) | PSNR: 23.57 -> 27.00 | Improved |
Closest papers by meaning
- Material Palette: Extraction of Materials from a Single Image (CVPR 2024 · similarity 0.90)
- TexGaussian: Generating High-quality PBR Material via Octree-based 3D Gaussian Splatting (CVPR 2025 · similarity 0.88)
- MAGE : Single Image to Material-Aware 3D via the Multi-View G-Buffer Estimation Model (CVPR 2025 · similarity 0.87)
- Intrinsic Image Diffusion for Indoor Single-view Material Estimation (CVPR 2024 · similarity 0.87)
- MaterialRefGS: Reflective Gaussian Splatting with Multi-view Consistent Material Inference (NeurIPS 2025 · similarity 0.87)
- Alchemist: Parametric Control of Material Properties with Diffusion Models (CVPR 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.
