Research map / Image generation / Super-resolution
Fast Image Super-Resolution via Consistency Rectified Flow
Jiaqi Xu, Wenbo Li, Haoze Sun, Fan Li, Zhixin Wang, Long Peng, Jingjing Ren, Haoran Yang, Xiaowei Hu, Renjing Pei, Pheng-Ann Heng
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 SR flow from LR instead of noise->HR flow, 1 step (ablation) | FlowSR · RealSR | PSNR: 24.82 -> 25.51 | Improved |
| use SR flow from LR instead of noised LR (ResShift-style), 1 step (ablation) | FlowSR · RealSR | PSNR: 24.67 -> 25.51 | Improved |
| use SR flow instead of noise->HR flow, 4 steps (ablation) | FlowSR · DRealSR | MUSIQ: 62.76 -> 65.05 | Improved |
| add plain consistency distillation L_cd (ablation) | SR flow, 1 step · DRealSR | MUSIQ: 62.53 -> 61.69 | Hurt |
| add HR-regularized consistency L_hrcd (ablation) | SR flow, 1 step · DRealSR | MUSIQ: 62.53 -> 64.77 | Improved |
| add HR regularization L_hr alone, without consistency (ablation) | SR flow, 1 step · DRealSR | MUSIQ: 62.53 -> 62.46 | No clear effect |
| add L_cd consistency distillation (ablation) | SR flow, 1 step · DRealSR | LPIPS: 0.2821 -> 0.2609 | Improved |
| use fast-slow (4) scheduling instead of N-interval (4) (ablation) | consistency SR flow · DRealSR | MUSIQ: 63.03 -> 64.77 | Improved |
| use fast-slow (4) instead of slow-only (1000) (ablation) | consistency SR flow · DRealSR | MUSIQ: 62.72 -> 64.77 | Improved |
| use 1 fast timestep instead of 4 (ablation) | fast-slow scheduling · DRealSR | MUSIQ: 64.77 -> 63.01 | Hurt |
| add HR-regularized consistency loss L_hrcd to full model (ablation) | FlowSR · DRealSR | LPIPS: 0.3013 -> 0.2975 | Improved |
| add adversarial loss L_adv (ablation) | FlowSR · DRealSR | MUSIQ: 65.29 -> 65.46 | Improved |
Closest papers by meaning
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · similarity 0.91)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.91)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.90)
- Efficient Conditional Diffusion Model with Probability Flow Sampling for Image Super-resolution (AAAI 2024 · similarity 0.89)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.89)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.89)
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.
