Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.08771 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908903923515392 |
|---|---|
| author | Li, Xiaohui Zhuang, Shaobin Cao, Shuo Yang, Yang Pu, Yuandong Qin, Qi Luo, Siqi Fu, Bin Liu, Yihao |
| author_facet | Li, Xiaohui Zhuang, Shaobin Cao, Shuo Yang, Yang Pu, Yuandong Qin, Qi Luo, Siqi Fu, Bin Liu, Yihao |
| contents | Generative models for Image Super-Resolution (SR) are increasingly powerful, yet their reliance on self-attention's quadratic complexity (O(N^2)) creates a major computational bottleneck. Linear Attention offers an O(N) solution, but its promise for photorealistic SR has remained largely untapped, historically hindered by a cascade of interrelated and previously unsolved challenges. This paper introduces LinearSR, a holistic framework that, for the first time, systematically overcomes these critical hurdles. Specifically, we resolve a fundamental, training instability that causes catastrophic model divergence using our novel "knee point"-based Early-Stopping Guided Fine-tuning (ESGF) strategy. Furthermore, we mitigate the classic perception-distortion trade-off with a dedicated SNR-based Mixture of Experts (MoE) architecture. Finally, we establish an effective and lightweight guidance paradigm, TAG, derived from our "precision-over-volume" principle. Our resulting LinearSR model simultaneously delivers state-of-the-art perceptual quality with exceptional efficiency. Its core diffusion forward pass (1-NFE) achieves SOTA-level speed, while its overall multi-step inference time remains highly competitive. This work provides the first robust methodology for applying Linear Attention in the photorealistic SR domain, establishing a foundational paradigm for future research in efficient generative super-resolution. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_08771 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LinearSR: Unlocking Linear Attention for Stable and Efficient Image Super-Resolution Li, Xiaohui Zhuang, Shaobin Cao, Shuo Yang, Yang Pu, Yuandong Qin, Qi Luo, Siqi Fu, Bin Liu, Yihao Computer Vision and Pattern Recognition Generative models for Image Super-Resolution (SR) are increasingly powerful, yet their reliance on self-attention's quadratic complexity (O(N^2)) creates a major computational bottleneck. Linear Attention offers an O(N) solution, but its promise for photorealistic SR has remained largely untapped, historically hindered by a cascade of interrelated and previously unsolved challenges. This paper introduces LinearSR, a holistic framework that, for the first time, systematically overcomes these critical hurdles. Specifically, we resolve a fundamental, training instability that causes catastrophic model divergence using our novel "knee point"-based Early-Stopping Guided Fine-tuning (ESGF) strategy. Furthermore, we mitigate the classic perception-distortion trade-off with a dedicated SNR-based Mixture of Experts (MoE) architecture. Finally, we establish an effective and lightweight guidance paradigm, TAG, derived from our "precision-over-volume" principle. Our resulting LinearSR model simultaneously delivers state-of-the-art perceptual quality with exceptional efficiency. Its core diffusion forward pass (1-NFE) achieves SOTA-level speed, while its overall multi-step inference time remains highly competitive. This work provides the first robust methodology for applying Linear Attention in the photorealistic SR domain, establishing a foundational paradigm for future research in efficient generative super-resolution. |
| title | LinearSR: Unlocking Linear Attention for Stable and Efficient Image Super-Resolution |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.08771 |