Research map / 3D vision / Face and shape
Shape, Light, and Material Decomposition from Images using Monte Carlo Rendering and Denoising
Jon Hasselgren, Nikolai Hofmann, Jacob Munkberg
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 Monte Carlo direct illumination with ray tracing, multiple importance sampling, and denoising instead of the simpler lighting pipeline | nvdiffrec-style inverse rendering · NeRFactor synthetic relighting | PSNR: 24.8~dB -> 26.0~dB | Improved |
| Use Monte Carlo direct illumination with ray tracing, multiple importance sampling, and denoising instead of the simpler lighting pipeline | nvdiffrec-style inverse rendering · NeRF synthetic relighting | PSNR: 23.3~dB -> 26.5~dB | Improved |
| Use Monte Carlo direct illumination with ray tracing, multiple importance sampling, and denoising instead of the simpler lighting pipeline | nvdiffrec-style inverse rendering · Our synthetic relighting | PSNR: 23.7~dB -> 27.1~dB | Improved |
| Use physically based material/light separation and regularization rather than the nvdiffrec pipeline | nvdiffrec-style inverse rendering · NeRFactor synthetic view interpolation | PSNR: 31.7~dB -> 29.6~dB | Hurt |
| Use physically based material/light separation and regularization rather than the nvdiffrec pipeline | nvdiffrec-style inverse rendering · NeRF synthetic view interpolation | PSNR: 30.4~dB -> 28.4~dB | Hurt |
| Use physically based material/light separation and regularization rather than the nvdiffrec pipeline | nvdiffrec-style inverse rendering · Real-world view interpolation | PSNR: 26.58~dB -> 25.29~dB | Hurt |
| Add denoising during optimization (ablation) | Monte Carlo inverse rendering · Porsche reconstruction and environment lighting | reconstruction quality | Improved |
| Use denoising at 8 spp rather than no denoising at 32 spp (ablation) | Monte Carlo inverse rendering · Scene reconstruction | reconstruction quality: 8~spp; 32~spp | Improved |
| Fine-tune denoiser weights jointly with scene parameters rather than keeping denoiser weights fixed (ablation) | Monte Carlo inverse rendering · Scene reconstruction | reconstruction quality | Hurt |
| Add the monochrome lighting regularizer (ablation) | Monte Carlo inverse rendering · Hotdog, NeRF synthetic | lighting/material separation | Improved |
| Ramp in the shadow contribution during early optimization instead of applying the full shadow term immediately (ablation) | Monte Carlo inverse rendering · Geometry reconstruction | geometry reconstruction | Improved |
| Enable shadow visibility gradients (ablation) | Joint shape, material, and lighting optimization · Three validation scenes | PSNR: 29.06 -> 29.15; 26.83 -> 27.03; 28.21 -> 27.83 | Mixed |
Closest papers by meaning
- Inverse Rendering using Multi-Bounce Path Tracing and Reservoir Sampling (ICLR 2025 · similarity 0.92)
- Multi-times Monte Carlo Rendering for Inter-reflection Reconstruction (NeurIPS 2024 · similarity 0.92)
- Eclipse: Disambiguating Illumination and Materials using Unintended Shadows (CVPR 2024 · similarity 0.91)
- MaterialRefGS: Reflective Gaussian Splatting with Multi-view Consistent Material Inference (NeurIPS 2025 · similarity 0.88)
- Factored-NeuS: Reconstructing Surfaces, Illumination, and Materials of Possibly Glossy Objects (CVPR 2025 · similarity 0.88)
- IDArb: Intrinsic Decomposition for Arbitrary Number of Input Views and Illuminations (ICLR 2025 · similarity 0.88)
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.
