Research map / Vision-language and multimodal / Cross-modal learning
UniLight: A Unified Representation for Lighting
Zitian Zhang, Iliyan Georgiev, Michael Fischer, Yannick Hold-Geoffroy, Jean-Francois Lalonde, Valentin Deschaintre
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 spherical-harmonics supervision (SH3) (ablation) | UniLight 8 tokens · Lighting retrieval (603 test samples) | R@1: 10.2 -> 24.9 | Improved |
| add spherical-harmonics supervision (SH3) (ablation) | UniLight 8 tokens · Lighting retrieval (603 test samples) | MRR: 0.215 -> 0.367 | Improved |
| raise SH degree from 1 to 3 (ablation) | UniLight 8 tokens · Lighting retrieval (603 test samples) | R@1: 20.7 -> 24.9 | Improved |
| raise SH degree from 3 to 5 (ablation) | UniLight 8 tokens · Lighting retrieval (603 test samples) | R@1: 24.9 -> 24.1 | Hurt |
| increase tokens from 1 to 8 (ablation) | UniLight SH3 · Lighting retrieval (603 test samples) | R@1: 23.8 -> 24.9 | Improved |
| increase tokens from 8 to 16 (ablation) | UniLight SH3 · Lighting retrieval (603 test samples) | R@1: 24.9 -> 26.5 | Improved |
| increase tokens from 4 to 8 (ablation) | UniLight SH3 · Lighting retrieval (603 test samples) | R@1: 25.1 -> 24.9 | No clear effect |
| use UniLight embedding instead of Qwen3-VL 2B embedding | retrieval · Lighting retrieval (603 test samples) | R@1: 8.9 -> 24.9 | Improved |
| use UniLight embedding instead of CLIP ViT-B/32 | retrieval · Lighting retrieval (603 test samples) | R@1: 2.6 -> 24.9 | Improved |
| use UniLight-conditioned diffusion instead of DiffusionLight-Turbo for env-map | env-map generation · HDR env-map renders | PSNR: 27.77 -> 28.85 | Improved |
| use UniLight-conditioned diffusion instead of DiffusionLight-Turbo for env-map | env-map generation · HDR env-map renders | LPIPS: 0.088 -> 0.079 | Improved |
Closest papers by meaning
- A Pre-convolved Representation for Plug-and-Play Neural Illumination Fields (AAAI 2024 · similarity 0.87)
- DreamLight: Towards Harmonious and Consistent Image Relighting (NeurIPS 2025 · similarity 0.87)
- LumiNet: Latent Intrinsics Meets Diffusion Models for Indoor Scene Relighting (CVPR 2025 · similarity 0.87)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.87)
- UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback (NeurIPS 2025 · similarity 0.86)
- Learning Adaptive Lighting via Channel-Aware Guidance (ICML 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.
