TOC-SR：基于任务最优紧凑扩散的图像超分辨率
计算机视觉与模式识别
2026-05-05 v1 人工智能
摘要
扩散模型最近在图像修复任务中展现出强大的性能，包括超分辨率。然而，其大型模型规模和迭代采样程序使得其在实际部署时计算成本较高。本文提出了TOC-SR框架，通过首先发现紧凑的扩散主干网络来构建高效的单步超分辨率模型。从十六通道潜在扩散模型开始，我们使用特征导向生成蒸馏构建参数高效的替代块，并通过epsilon约束贝叶斯优化进行架构发现，以最小化模型复杂度同时保持生成保真度。 resulting compact diffusion backbone achieves a 6.6x reduction in parameters and a 2.8x reduction in GMACs compared to the expanded diffusion model. We then adapt this backbone for super-resolution and distill the diffusion process into a single-step generator. Experiments demonstrate that the proposed approach enables efficient super-resolution while maintaining strong reconstruction quality.
引用
@article{arxiv.2605.02767,
title = {TOC-SR: Task-Optimal Compact diffusion for Image Super Resolution},
author = {Sowmya Vajrala and Akshay Bankar and Manjunath Arveti and Shreyas Pandith and Sravanth Kodavanti and Subhajit Sanyal and Amit Unde and Srinivas Soumitri Miriyala},
journal= {arXiv preprint arXiv:2605.02767},
year = {2026}
}