Research map / Image generation / Super-resolution
Feature Representation Matters: End-to-End Learning for Reference-based Image Super-resolution
Yanchun Xie, Jimin Xiao, Mingjie Sun, Chao Yao, Kaizhu Huang
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 trainable end-to-end SR feature encoder instead of ImageNet VGG features (ablation) | E2ENT2 · CUFED5 | PSNR: 22.85 -> 24.01 | Improved |
| use end-to-end trained encoder instead of frozen pre-trained SISR encoder (ablation) | E2ENT2 · CUFED5 | PSNR: 23.46 -> 24.01 | Improved |
| use end-to-end trained encoder instead of VGG, feature distance (ablation) | E2ENT2 · CUFED5 | feature distance (lower better): 106.77 -> 25.77 | Improved |
| allocate gradient across all three features instead of only the LR input feature (ablation) | E2ENT2 · CUFED5 | PSNR: 23.75 -> 24.01 | Improved |
| allocate gradient across all three features instead of only the downsampled reference feature (ablation) | E2ENT2 · CUFED5 | PSNR: 23.67 -> 24.01 | Improved |
| allocate gradient across all three features instead of only the HR reference feature (ablation) | E2ENT2 · CUFED5 | PSNR: 23.83 -> 24.01 | Improved |
| replace L1 reconstruction loss with MSE loss (ablation) | E2ENT2 · CUFED5 | PSNR: 24.01 -> 24.24 | Improved |
| replace L1 reconstruction loss with MSE loss (ablation) | E2ENT2 · SUN Hays | PSNR: 28.13 -> 28.50 | Improved |
| add reference image with end-to-end match and swap | Ours-SISR · CUFED5 | PSNR: 23.75 -> 24.01 | Improved |
| add reference image with end-to-end match and swap | Ours-SISR · SUN Hays | PSNR: 26.72 -> 28.13 | Improved |
| use E2ENT2 instead of SRNTT texture transfer | SRNTT · CUFED5 and SUN Hays | PSNR: +0.6 dB on CUFED5, +1.71 dB on SUN Hays | Improved |
| use E2ENT2 instead of SRNTT on stereo data | SRNTT · Flickr1024 test | PSNR: 22.02 -> 22.89 | Improved |
Closest papers by meaning
- Image Super-Resolution by Neural Texture Transfer (CVPR 2019 · similarity 0.93)
- Learning Texture Transformer Network for Image Super-Resolution (CVPR 2020 · similarity 0.92)
- Task Decoupled Framework for Reference-Based Super-Resolution (CVPR 2022 · similarity 0.91)
- Attention-Based Multi-Reference Learning for Image Super-Resolution (ICCV 2021 · similarity 0.91)
- Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss (CVPR 2024 · similarity 0.87)
- Auto-Encoded Supervision for Perceptual Image Super-Resolution (CVPR 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.
