Research map / 3D vision / Gaussian splatting
Flash Cache: Reducing Bias in Radiance Cache Based Inverse Rendering
Benjamin Attal, Dor Verbin, Ben Mildenhall, Peter Hedman, Jonathan T Barron, Matthew O'Toole, Pratul Srinivasan
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 fast-cache control variate (ablation) | Flash Cache · TensoIR-Synthetic (single light) | NVS PSNR: 34.629 -> 34.915 | Improved |
| add vMF occlusion-aware importance sampler (ablation) | Flash Cache · TensoIR-Synthetic (single light) | Albedo PSNR: 30.237 -> 30.329 | Improved |
| add vMF importance sampler on top of no-variate model (ablation) | Flash Cache · TensoIR-Synthetic (single light) | NVS PSNR: 34.653 -> 34.629 | No clear effect |
| use unbiased single-sample volume rendering estimator instead of median depth sampling (ablation) | Flash Cache · TensoIR-Synthetic (single light) | Normal MAE: 3.491 -> 3.352 | Improved |
| use unbiased volume rendering estimator instead of median depth (ablation) | Flash Cache · TensoIR-Synthetic (single light) | NVS PSNR: 34.249 -> 34.653 | Improved |
| increase samples along ray K from 1 to 8 (ablation) | Flash Cache · TensoIR-Synthetic (single light) | Albedo PSNR: 30.345 -> 30.059 | Hurt |
| use unbiased radiance-cache inverse rendering vs TensoIR | Flash Cache · TensoIR-Synthetic | Normal MAE: 4.100 -> 3.355 | Improved |
| Flash Cache vs TensoIR on novel view synthesis | Flash Cache · TensoIR-Synthetic | NVS PSNR: 35.088 -> 34.908 | No clear effect |
| Flash Cache vs TensoIR on specular scenes | Flash Cache · Open Illumination (specular scenes) | Relighting PSNR: 26.955 -> 27.810 | Improved |
| Flash Cache vs TensoIR on diffuse scenes | Flash Cache · Open Illumination (diffuse scenes) | Relighting PSNR: 30.932 -> 29.679 | Hurt |
| use fast cache instead of NeRF cache for secondary rays | Flash Cache · secondary ray cost | seconds per million rays: 1.574 -> 0.128 | Improved |
Closest papers by meaning
- Inverse Rendering using Multi-Bounce Path Tracing and Reservoir Sampling (ICLR 2025 · similarity 0.90)
- Neural Inverse Rendering from Propagating Light (CVPR 2025 · similarity 0.90)
- NeFII: Inverse Rendering for Reflectance Decomposition With Near-Field Indirect Illumination (CVPR 2023 · similarity 0.89)
- GI-GS: Global Illumination Decomposition on Gaussian Splatting for Inverse Rendering (ICLR 2025 · similarity 0.88)
- Multi-times Monte Carlo Rendering for Inter-reflection Reconstruction (NeurIPS 2024 · similarity 0.87)
- Inverse Rendering of Glossy Objects via the Neural Plenoptic Function and Radiance Fields (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.
