Research map / 3D vision / Depth and stereo
CroCoDiLight: Repurposing Cross-View Completion Encoders for Relighting
Alistair J Foggin, William A P Smith
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 CroCo pre-trained encoder instead of simple encoder trained from scratch (ablation) | CroCoDiLight · IIW | WHDR (%): 19.1% -> 15.4% | Improved |
| use CroCo pre-trained weights instead of same architecture from scratch (ablation) | CroCoDiLight · IIW | WHDR (%): 27.7% -> 15.4% | Improved |
| use CroCo pre-trained encoder instead of simple encoder from scratch (ablation) | CroCoDiLight · ISTD+ | PSNR: 29.71 -> 30.17 | Improved |
| use CroCo pre-trained weights instead of CroCo architecture from scratch (ablation) | CroCoDiLight · ISTD+ | PSNR: 25.34 -> 30.17 | Improved |
| interpolate in lighting latent space instead of image space | timelapse upsampling · Clock Shadows | FloLPIPS: 0.309 -> 0.286 | Improved |
| interpolate in lighting latent space instead of image space | timelapse upsampling · Day-night Cycle | FloLPIPS: 0.041 -> 0.043 | Hurt |
| interpolate in lighting latent space instead of image space | timelapse upsampling · Indoor Shadows | FloLPIPS: 0.950 -> 0.923 | Improved |
| add the +0.5 RGB shift at evaluation | CroCoDiLight · IIW | WHDR (%): 15.4% -> 14.3% | Improved |
| add synthetic INS dataset to training of I and R (ablation) | CroCoDiLight · ISTD+ | PSNR: 30.17 -> 31.23 | Improved |
| add synthetic INS dataset to training of I and R (ablation) | CroCoDiLight · IIW albedo estimation | WHDR (%): 15.4% -> 17.6% | Hurt |
| add INS dataset to training of S as well (ablation) | CroCoDiLight · SRD | PSNR: 30.55 -> 29.38 | Hurt |
| use the oracle shadow-free latent instead of learned mapping S (ablation) | CroCoDiLight · ISTD+ | PSNR: 30.17 -> 33.41 | Improved |
Closest papers by meaning
- CroCo v2: Improved Cross-view Completion Pre-training for Stereo Matching and Optical Flow (ICCV 2023 · similarity 0.91)
- Cross-View Completion Models are Zero-shot Correspondence Estimators (CVPR 2025 · similarity 0.88)
- UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback (NeurIPS 2025 · similarity 0.86)
- ReCap: Better Gaussian Relighting with Cross-Environment Captures (CVPR 2025 · similarity 0.85)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.85)
- MetaGS: A Meta-Learned Gaussian-Phong Model for Out-of-Distribution 3D Scene Relighting (NeurIPS 2025 · similarity 0.85)
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.
