Research map / Image generation / Super-resolution
Fourier Space Losses for Efficient Perceptual Image Super-Resolution
Dario Fuoli, Luc Van Gool, Radu Timofte
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 |
|---|---|---|---|
| use Fourier supervision loss instead of L1 (ablation) | IMDN · DIV2K val (loss ablation, 4x) | LPIPS: 0.270 -> 0.189 | Improved |
| use Fourier supervision loss instead of L1 (ablation) | IMDN · DIV2K val (loss ablation, 4x) | PSNR: 30.56 -> 29.53 | Hurt |
| add Fourier supervision loss to L1 (ablation) | IMDN · DIV2K val (loss ablation, 4x) | FID: 22.91 -> 21.96 | Improved |
| replace ESRGAN spatial losses with Fourier losses (ablation) | IMDN · DIV2K val (loss ablation, 4x) | PSNR: 27.94 -> 29.06 | Improved |
| replace ESRGAN spatial losses with Fourier losses (ablation) | IMDN · DIV2K val (loss ablation, 4x) | LPIPS: 0.131 -> 0.129 | No clear effect |
| add Fourier losses on top of spatial losses (ablation) | IMDN · DIV2K val (loss ablation, 4x) | FID: 17.07 -> 15.88 | Improved |
| add Fourier GAN loss (ablation) | IMDN · DIV2K val (loss ablation, 4x) | FID: 16.94 -> 15.88 | Improved |
| add spatial GAN loss alongside Fourier GAN (ablation) | IMDN · DIV2K val (loss ablation, 4x) | FID: 17.90 -> 15.88 | Improved |
| add Fourier losses to ESRGAN generator (ablation) | ESRGAN · DIV2K val (loss ablation, 4x) | FID: 15.37 -> 14.80 | Improved |
| add Fourier losses to ESRGAN generator | ESRGAN · Urban100 | FID: 25.50 -> 24.07 | Improved |
| add Fourier losses to ESRGAN generator | ESRGAN · BSD100 | FID: 60.42 -> 57.90 | Improved |
| use Fourier losses instead of WaveletSRNet losses | IMDN · DIV2K val (comparison with state of the art) | FID: 19.80 -> 15.72 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SRFlow — SRFlow: Learning the Super-Resolution Space with Normalizing Flow (ECCV 2020) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Auto-Encoded Supervision for Perceptual Image Super-Resolution (CVPR 2025 · similarity 0.89)
- Perception-Oriented Single Image Super-Resolution Using Optimal Objective Estimation (CVPR 2023 · similarity 0.89)
- Training Generative Image Super-Resolution Models by Wavelet-Domain Losses Enables Better Control of Artifacts (CVPR 2024 · similarity 0.88)
- SROBB: Targeted Perceptual Loss for Single Image Super-Resolution (ICCV 2019 · similarity 0.88)
- Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss (CVPR 2024 · similarity 0.87)
- Augmenting Perceptual Super-Resolution via Image Quality Predictors (CVPR 2025 · similarity 0.87)
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.
