Research map / Image generation / Image restoration
Light Up Your Face: A Physically Consistent Dataset and Diffusion Model for Face Fill-Light Enhancement
Jue Gong, Zihan Zhou, Jingkai Wang, Xiaohong Liu, Yulun Zhang, Xiaokang Yang
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 |
|---|---|---|---|
| pretrain PALP with both planar irradiance map and direction field targets (ablation) | FiLitDiff · LYF-Val | SSIM: 0.8748 -> 0.8773 | Improved |
| pretrain PALP with both targets (ablation) | FiLitDiff · LYF-Val | LIQE: 3.7798 -> 3.8121 | Improved |
| pretrain PALP with irradiance map only (no direction field) (ablation) | FiLitDiff · LYF-Val | SSIM: 0.8748 -> 0.8731 | No clear effect |
| pretrain PALP with direction field only (ablation) | FiLitDiff · LYF-Val | CLIPIQA: 0.5839 -> 0.5839 | No clear effect |
| pretrain PALP separately before joint training instead of from scratch with the diffusion model (ablation) | FiLitDiff · LYF-Val | DISTS: 0.0763 -> 0.0756 | Improved |
| add MSSWD color loss to DISTS loss (ablation) | FiLitDiff · LYF-Val | DISTS: 0.0769 -> 0.0756 | Improved |
| add MSSWD color loss to DISTS loss (ablation) | FiLitDiff · LYF-Val | SSIM: 0.8649 -> 0.8773 | Improved |
| use LPIPS instead of DISTS with MSSWD (ablation) | FiLitDiff · LYF-Val | DISTS: 0.0756 -> 0.0914 | Hurt |
| use LPIPS instead of DISTS with MSSWD (ablation) | FiLitDiff · LYF-Val | CLIPIQA: 0.5849 -> 0.5419 | Hurt |
| fine-tune physics-aware lighting model FiLitDiff instead of using DPR, SMFR, IC-Light baselines | FiLitDiff · LYF-Val | PSNR: 16.67 -> 25.65 | Improved |
| use prompt-based Qwen-Image-Edit instead of FiLitDiff | Qwen-Image-Edit · LYF-Val | TOPIQ: 0.7490 vs 0.7681 | Mixed |
| larger training target strength coefficient than 0.6 | FiLitDiff · training target | perceptual distance vs fidelity | Mixed |
Closest papers by meaning
- SynthLight: Portrait Relighting with Diffusion Model by Learning to Re-render Synthetic Faces (CVPR 2025 · similarity 0.88)
- Lux Post Facto: Learning Portrait Performance Relighting with Conditional Video Diffusion and a Hybrid Dataset (CVPR 2025 · similarity 0.88)
- Learning to Decouple the Lights for 3D Face Texture Modeling (NeurIPS 2024 · similarity 0.87)
- Low-Light Face Super-resolution via Illumination, Structure, and Texture Associated Representation (AAAI 2024 · similarity 0.86)
- Zero-Reference Low-Light Enhancement via Physical Quadruple Priors (CVPR 2024 · similarity 0.86)
- DreamLight: Towards Harmonious and Consistent Image 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.
