Research map / Image generation / Super-resolution
One-Step Residual Shifting Diffusion for Image Super-Resolution via Distillation
Daniil Selikhanovych, David Li, Aleksei Leonov, Nikita Gushchin, Sergei Kushneriuk, Alexander Filippov, Evgeny Burnaev, Iaroslav Sergeevich Koshelev, Alexander Korotin
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 |
|---|---|---|---|
| train RSD with N=4 timesteps instead of single-step N=1 (ablation) | RSD · RealSR | LPIPS: 0.4052 -> 0.3552 | Improved |
| increase timesteps N from 4 to 15 (ablation) | RSD · RealSR | PSNR / CLIPIQA: PSNR 24.92 -> 25.91; CLIPIQA 0.7518 -> 0.6857 | Mixed |
| increase timesteps N from 1 to 8 (ablation) | RSD · RealSR | MUSIQ: 64.290 -> 66.445 | Improved |
| add both supervised losses (lambda_1, lambda_2 nonzero) to RSD (ablation) | RSD · RealSR | PSNR / LPIPS: PSNR 24.92 -> 25.91; LPIPS 0.3552 -> 0.2726 | Improved |
| add both supervised losses to RSD (ablation) | RSD · RealSR | CLIPIQA / MUSIQ: CLIPIQA 0.7518 -> 0.7060; MUSIQ 66.430 -> 65.860 | Hurt |
| enable only lambda_2 supervised loss (ablation) | RSD · RealSR | MUSIQ: 66.430 -> 67.615 | Improved |
| distill ResShift with RSD loss instead of SinSR | ResShift teacher · RealSR | LPIPS: 0.365 -> 0.273 | Improved |
| use RSD loss instead of VSD loss for ResShift distillation | ResShift one-step student · RealSR | LPIPS / PSNR: LPIPS 0.466 -> 0.273; PSNR 23.96 -> 25.91 | Improved |
| RSD versus SinSR training/inference efficiency | RSD · Complexity | GPU memory / training time: 539 MB vs 570 MB; 5 / 4 A100 vs 60 / 1 A100 hours | Improved |
| compare RSD one-step with OSEDiff on DIV2K-Val | RSD · DIV2K-Val | PSNR: 23.72 -> 23.91 (RSD) but vs CTMSR 24.88 | Hurt |
Closest papers by meaning
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.93)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.93)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.92)
- Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator (NeurIPS 2025 · similarity 0.92)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.92)
- ResDiff: Combining CNN and Diffusion Model for Image Super-resolution (AAAI 2024 · similarity 0.91)
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.
