Research map / Image generation / Super-resolution
Allo{SR}2: Rectifying One-Step Super-Resolution to Stay Real via Allomorphic Generative Flows
Zihan Wang, XUDONG HUANG, Junbo Qiao, WEI LI, jie hu, Xinghao Chen, Shaohui Lin
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 SNR-selected initial timestep t* instead of t=999 (ablation) | AlloSR2 · RealSR | PSNR: 24.16 -> 24.97 | Improved |
| add SNR-selected initial timestep t* instead of t=999 (ablation) | AlloSR2 · RealSR | MANIQA: 0.6713 -> 0.6725 | Improved |
| add SNR-selected initial timestep t* instead of t=1 (ablation) | AlloSR2 · RealSR | MUSIQ: 68.90 -> 70.11 | Improved |
| use t* instead of t=1 (ablation) | AlloSR2 · RealSR | PSNR: 24.74 -> 24.97 | No clear effect |
| add SNR-selected initial timestep t* (ablation) | AlloSR2 · RealSR | training iterations to converge: 9k -> 7k | Improved |
| add FATC loss (ablation) | AlloSR2 · DRealSR | LPIPS: 0.3106 -> 0.2989 | Improved |
| add FATC loss (ablation) | AlloSR2 · DRealSR | CLIPIQA: 0.6357 -> 0.6883 | Improved |
| add FATC loss (ablation) | AlloSR2 · DRealSR | FID: 121.75 -> 132.48 | Hurt |
| add ATM loss (ablation) | AlloSR2 · DRealSR | FID: 139.74 -> 132.48 | Improved |
| add ATM loss (ablation) | AlloSR2 · DRealSR | MUSIQ: 62.77 -> 66.95 | Improved |
| add ATM loss (ablation) | AlloSR2 · DRealSR | PSNR: 28.68 -> 27.95 | Hurt |
| train and test at 1024x1024 instead of 512x512 | AlloSR2 · RealLQ250 | MANIQA: 0.6303 -> 0.6497 | Improved |
Closest papers by meaning
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.89)
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · similarity 0.88)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.88)
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · similarity 0.88)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.88)
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · 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.
