Research map / Image generation / Image restoration
CuDi: Curve Distillation for Efficient and Controllable Exposure Adjustment
Chongyi Li, Chunle Guo, Ruicheng Feng, Shangchen Zhou, Chen Change Loy
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 curve distillation (student branch) to teacher branch (ablation) | CuDi · 2K image | GPU runtime: 1.6s -> 4.9ms | Improved |
| add curve distillation (ablation) | CuDi · model size | parameters: 4.7M -> 3K | Improved |
| add curve distillation (ablation) | CuDi · 2K image | FLOPs: 10.4T -> 0.5G | Improved |
| replace iterative LE8 with tangent-line linear operation (ablation) | CuDi · 8K image | GPU runtime: 58.1ms -> 3.6ms | Improved |
| replace iterative LE8 with tangent-line linear operation (ablation) | CuDi · 16K image | GPU runtime: 0.2s -> 0.01s | Improved |
| distill teacher into student (ablation) | CuDi · 100 test images | SSIM vs teacher: 0.952 SSIM, 0.998 PCC | Improved |
| train student directly without distillation (ablation) | CuDi · visual comparison | visual quality | Hurt |
| replace tangent line with a 3-layer CNN in distillation (ablation) | CuDi · visual comparison | visual quality | Hurt |
| jointly train teacher and student instead of sequentially (ablation) | CuDi · visual comparison | visual quality | Hurt |
| approximate LE8 instead of using only LE1 or LE2 (ablation) | CuDi · visual comparison | exposure adjustment strength | Improved |
| use CuDi instead of Zero-DCE++ (ablation) | Zero-DCE++ · 4096x4096 | GPU runtime: 34.37 -> 14.77 | Improved |
Closest papers by meaning
- Continuous Exposure Learning for Low-light Image Enhancement using Neural ODEs (ICLR 2025 · similarity 0.88)
- AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025 · similarity 0.85)
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.84)
- Real-Time Exposure Correction via Collaborative Transformations and Adaptive Sampling (CVPR 2024 · similarity 0.84)
- Efficient Diffusion as Low Light Enhancer (CVPR 2025 · similarity 0.84)
- ECMamba: Consolidating Selective State Space Model with Retinex Guidance for Efficient Multiple Exposure Correction (NeurIPS 2024 · similarity 0.83)
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.
