Research map / Image generation / Autoregressive and latent generation
Relighting as a Probe of Visual Priors via Augmented Latent Intrinsics
Xiaoyan Xing, Xiao Zhang, Sezer Karaoglu, Theo Gevers, Anand Bhattad
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 |
|---|---|---|---|
| augment latent intrinsics with RADIOv2.5 features (ablation) | ALI · MIIW (Stage I&II) | PSNR: 17.5463 -> 18.3426 | Improved |
| augment with MAE features instead of none (ablation) | ALI · MIIW (Stage I&II) | PSNR: 17.5463 -> 17.9861 | Improved |
| augment with DINOv2 features instead of none (ablation) | ALI · MIIW (Stage I&II) | PSNR: 17.5463 -> 15.7945 | Hurt |
| augment with CLIP features instead of none (ablation) | ALI · MIIW (Stage I&II) | PSNR: 17.5463 -> 16.1333 | Hurt |
| upgrade DINOv2 to DINOv3 (denser features) (ablation) | ALI · MIIW (Stage I&II) | PSNR: 15.7945 -> 16.1510 | Improved |
| add Stage II decoder alignment after Stage I with MAE features (ablation) | ALI · MIIW | PSNR: 14.2381 -> 17.9861 | Improved |
| add Stage II decoder alignment after Stage I with RADIOv2.5 (ablation) | ALI · MIIW | SSIM: 0.5609 -> 0.5958 | Improved |
| add Stage I semantic augmentation (ablation) | ALI · MIIW Glossy regions | SSIM: 0.5720 -> 0.5988 | Improved |
| add Stage I semantic augmentation (ablation) | ALI · MIIW Metallic regions | SSIM: 0.4164 -> 0.4608 | Improved |
| add Stage III self-refinement after Stage I&II (ablation) | ALI · MIIW Glossy regions | SSIM: 0.6291 -> 0.6196 | No clear effect |
| add Stage III self-refinement (ablation) | ALI · user study, lighting alignment | preference score: 0.21 -> 0.75 | Improved |
| use ALI instead of LumiNet | relighting model · MIIW cross-scene (color-corrected) | SSIM: 0.527 -> 0.553 | Improved |
Closest papers by meaning
- Latent Intrinsics Emerge from Training to Relight (NeurIPS 2024 · similarity 0.90)
- LumiNet: Latent Intrinsics Meets Diffusion Models for Indoor Scene Relighting (CVPR 2025 · similarity 0.89)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.89)
- DreamLight: Towards Harmonious and Consistent Image Relighting (NeurIPS 2025 · similarity 0.89)
- UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback (NeurIPS 2025 · similarity 0.88)
- ReCap: Better Gaussian Relighting with Cross-Environment Captures (CVPR 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.
