Research map / Image generation / Super-resolution
TEASR: Training-Efficient Any-Step Diffusion Transformer for Real-World Image Super-Resolution
Xiang Gao, Chenxin Zhu, Yushun Fang, Qiang Hu, Xiaoyun Zhang
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 Decoupled Timestep Condition (DTC) (ablation) | TEASR dual-branch DiT · RealSR (1-step) | PSNR: 24.93 -> 25.52 | Improved |
| add DTC (ablation) | TEASR dual-branch DiT · RealSR (1-step) | FID: 114.49 -> 96.11 | Improved |
| add DTC (ablation) | TEASR dual-branch DiT · RealSR (1-step) | NIQE: 4.84 -> 6.72 | Hurt |
| add Time-Aware Rectification (TAR) on top of DTC (ablation) | TEASR · RealSR (1-step) | NIQE: 6.72 -> 5.43 | Improved |
| add TAR on top of DTC (ablation) | TEASR · RealSR (1-step) | PSNR: 25.52 -> 24.83 | Hurt |
| add TAR on top of DTC (ablation) | TEASR · RealSR (1-step) | MUSIQ: 66.39 -> 68.50 | Improved |
| use TAR weight (1-s)^2/s instead of constant weight 1 (ablation) | TEASR · RealSR (1-step) | MUSIQ: 66.39 -> 68.50 | Improved |
| increase sampling steps from 1 to 30 | TEASR · RealSR | PSNR: 24.83 -> 23.61 | Hurt |
| increase sampling steps from 1 to 30 | TEASR · RealSR | FID: 103.97 -> 118.41 | Hurt |
| increase sampling steps from 1 to 30 | TEASR · RealSR | NIQE: 5.4334 -> 5.2327 | Improved |
| use TEASR single-model SAD training instead of OSEDiff multi-model training | TEASR · Training memory | memory increase over LoRA baseline: 395% -> 15.6% | Improved |
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 0.92)
- Effective Diffusion Transformer Architecture for Image Super-Resolution (AAAI 2025 · similarity 0.92)
- Adversarial Diffusion Compression for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.91)
- Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator (NeurIPS 2025 · similarity 0.91)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 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.
