Research map / Image generation / Image restoration
CGIntrinsics: Better Intrinsic Image Decomposition through Physically-Based Rendering
Zhengqi Li, Noah Snavely
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 |
|---|---|---|---|
| render with high-quality BDPT and simple tone mapping (CGI) instead of PBRS renderings (ablation) | CGI-trained network · IIW test | WHDR: 26.1% -> 17.8% | Improved |
| compute scale-invariant MSE in linear domain instead of log domain (ablation) | CGI-trained network · IIW test | WHDR: 22.7% -> 17.8% | Improved |
| add multi-scale gradient matching loss (ablation) | CGI-trained network · IIW test | WHDR: 19.7% -> 17.8% | Improved |
| add ordinal loss derived from CGI ground truth (ablation) | CGI-trained network · IIW test | WHDR: 19.9% -> 17.8% | Improved |
| add reflectance smoothness loss (ablation) | network trained on All · IIW test | WHDR: 16.1% -> 15.5% | Improved |
| add augmented IIW annotations to CGI training (ablation) | Ours · IIW test | WHDR: 17.8% -> 16.2% | Improved |
| add SAW data to CGI+IIW(A) training | Ours · IIW test | WHDR: 16.2% -> 15.5% | Improved |
| apply guided filter post-processing | Ours (CGI) · IIW test | WHDR: 17.8% -> 17.1% | Improved |
| add SAW data (with SAW losses) to CGI+IIW(A) | Ours · SAW test (challenge AP) | AP%: 94.69 -> 97.93 | Improved |
| add shading smoothness loss (ablation) | network trained on CGI+IIW(A)+SAW · SAW test (challenge AP) | AP%: 94.87 -> 97.93 | Improved |
| add gradient matching loss (ablation) | CGI-trained network · SAW test (challenge AP) | AP%: 93.74 -> 94.05 | Improved |
| render with CGI pipeline instead of PBRS (ablation) | CGI-trained network · SAW test (challenge AP) | AP%: 87.09 -> 94.05 | Improved |
Closest papers by meaning
- Revisiting Deep Intrinsic Image Decompositions (CVPR 2018 · similarity 0.91)
- Single Image Intrinsic Decomposition without a Single Intrinsic Image (ECCV 2018 · similarity 0.89)
- Self-Supervised Intrinsic Image Decomposition (NeurIPS 2017 · similarity 0.88)
- BLURD: Benchmarking and Learning using a Unified Rendering and Diffusion Model (NeurIPS 2024 · similarity 0.86)
- FerretNet: Efficient Synthetic Image Detection via Local Pixel Dependencies (NeurIPS 2025 · similarity 0.86)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.86)
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.
