Research map / Image generation / GANs
Deep Photo Enhancer: Unpaired Learning for Image Enhancement From Photographs With GANs
Yu-Sheng Chen, Yu-Ching Wang, Man-Hsin Kao, Yung-Yu Chuang
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 global features to U-Net (ablation) | U-Net for MIT-Adobe 5K photographer retouching prediction · MIT-Adobe 5K test | PSNR: 22.13 -> 23.80 | Improved |
| Add global features to U-Net (ablation) | U-Net for MIT-Adobe 5K photographer retouching prediction · MIT-Adobe 5K test | SSIM: 0.879 -> 0.900 | Improved |
| Use global U-Net instead of U-Net (ablation) | local Laplacian approximation · MIT-Adobe 5K test | PSNR: 31.06 -> 33.93 | Improved |
| Replace WGAN-GP gradient penalty with the adaptive A-WGAN penalty and weighting (ablation) | One-way GAN enhancement · MIT-Adobe 5K test | PSNR: 21.42 -> 22.17 | Improved |
| Vary the WGAN-GP gradient-penalty weight λ (ablation) | WGAN-GP enhancement · MIT-Adobe 5K test | PSNR: 21.42; 21.19; 20.78 | Mixed |
| Use a two-way GAN instead of a one-way GAN (ablation) | WGAN-GP enhancement · MIT-Adobe 5K test | PSNR: 21.42 -> 21.96 | Improved |
| Use a two-way GAN instead of a one-way GAN (ablation) | A-WGAN enhancement · MIT-Adobe 5K test | PSNR: 22.17 -> 22.21 | Improved |
| Give the two generators individual batch-normalization layers (ablation) | 2-way WGAN-GP · MIT-Adobe 5K test | PSNR: 21.96 -> 22.27 | Improved |
| Give the two generators individual batch-normalization layers (ablation) | 2-way A-WGAN · MIT-Adobe 5K test | PSNR: 22.21 -> 22.37 | Improved |
| Use A-WGAN's adaptive weighting rather than fixed WGAN-GP weighting (ablation) | GAN training on the swissroll distribution · swissroll | convergence | Improved |
| Train the image enhancer using the HDR photograph set rather than MIT-Adobe 5K retouching labels | Deep Photo Enhancer · photographic image enhancement | user preference: 81% (323 among 400) | Improved |
| Remove individual batch normalization (ablation) | HDR-trained 2-way GAN enhancer · photographic image enhancement | visual color rendition | Hurt |
Closest papers by meaning
- Unpaired Image Enhancement Featuring Reinforcement-Learning-Controlled Image Editing Software (AAAI 2020 · similarity 0.90)
- Global and Local Enhancement Networks for Paired and Unpaired Image Enhancement (ECCV 2020 · similarity 0.88)
- Neural Photo Editing with Introspective Adversarial Networks (ICLR 2017 · similarity 0.87)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.87)
- Linearly Controllable GAN: Unsupervised Feature Categorization and Decomposition for Image Generation and Manipulation (ECCV 2024 · similarity 0.85)
- Rectified Wasserstein Generative Adversarial Networks for Perceptual Image Restoration (TPAMI 2022 · 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.
