⚠️ 以下所有内容总结都来自于 大语言模型的能力，如有错误，仅供参考，谨慎使用
🔴 请注意：千万不要用于严肃的学术场景，只能用于论文阅读前的初筛！
💗 如果您觉得我们的项目对您有帮助 ChatPaperFree ，还请您给我们一些鼓励！⭐️ HuggingFace免费体验
2025-11-25 更新
EatGAN: An Edge-Attention Guided Generative Adversarial Network for Single Image Super-Resolution
Authors:Penghao Rao, Tieyong Zeng
Single-image super-resolution (SISR) is an important task in image processing, aiming to enhance the resolution of imaging systems. Recently, SISR has made a significant leap and achieved promising results with deep learning. GAN-based models stand out among all the deep learning models because of their excellent performance in perceiving quality. However, it is rather difficult for them to reconstruct realistic high-frequency details and achieve stable training. To solve these issues, we introduce an Edge-Attention guided Generative Adversarial Network (EatGAN), the first GAN-based SISR model that simultaneously leverages edge priors both explicitly and implicitly inside the generator, which (i) proposes a Normalized Edge Attention (NEA) mechanism based on channel-affine and spatial gating that transforms edge prior into lightweight, learnable modulation parameters and injects and fuses them multiple times in a (ii) edge-guided hybrid residual block, which progressively enforces structural consistency across scales; and (iii) a composite generator objective combining pixel, perceptual, edge-gradient, and adversarial terms. Experiments show consistent state-of-the-art across distortion-oriented benchmarks and perception oriented benchmarks. Notably, our model achieves 40.87 dB and 0.073 (LPIPS) on Manga 109, which indicates that reframing image priors from passive guidance into a controllable modulation primitive for generators can chart a practical path toward trustworthy, high-fidelity Super-Resolution.
单图像超分辨率（SISR）是图像处理中的一项重要任务，旨在提高成像系统的分辨率。最近，SISR在深度学习领域取得了重大突破，并取得了有前景的结果。基于GAN的模型在所有深度学习模型中脱颖而出，因为它们在感知质量方面表现出卓越的性能。然而，它们很难重建逼真的高频细节，并且实现稳定的训练。为了解决这些问题，我们引入了Edge-Attention引导生成对抗网络（EatGAN），这是第一个基于GAN的SISR模型，该模型在生成器中同时显式利用边缘先验和隐式利用边缘先验。它（i）提出了一种基于通道仿射和空间门控的归一化边缘注意力（NEA）机制，该机制将边缘先验转换为轻量级、可学习的调制参数，并在（ii）边缘引导混合残差块中多次注入和融合这些参数，该块逐步强制执行跨尺度的结构一致性；（iii）复合生成器目标，结合像素、感知、边缘梯度和对抗性术语。实验表明，在面向失真的基准测试和面向感知的基准测试中，我们的模型均表现出卓越的性能。值得注意的是，我们的模型在Manga 109上实现了40.87 dB和0.073（LPIPS），这表明将图像先验从被动指导重新定义为可控的调制原始数据，可以为生成器开辟一条实现可靠、高保真超分辨率的实际路径。
论文及项目相关链接
PDF 17 pages (8 pages of main text + 3 pages of reference + 6 pages of supplementary material)
Summary
基于边缘先验的EatGAN模型在单图像超分辨率任务中表现出卓越性能。它通过引入边缘引导混合残差块和复合生成器目标，解决了重建高频细节和稳定训练的问题。实验表明，该模型在面向失真的基准测试和面向感知的基准测试中均表现出一致的最佳性能。
Key Takeaways
- 单图像超分辨率（SISR）是图像处理中的一项重要任务，旨在提高成像系统的分辨率。
- 基于GAN的模型在感知质量方面表现出卓越性能，但重建高频细节和稳定训练方面存在挑战。
- EatGAN是首个同时利用边缘先验的GAN-based SISR模型，该模型引入了Normalized Edge Attention（NEA）机制。
- NEA机制基于通道亲和性和空间门控，将边缘先验转换为轻量级、可学习的调制参数，并在边缘引导的混合残差块中多次注入和融合它们。
- Edge-guided hybrid residual block逐步强制跨尺度的结构一致性。
- 复合生成器目标结合了像素、感知、边缘梯度和对抗性术语。
- 实验表明，EatGAN在面向失真的基准测试和面向感知的基准测试中均达到最新状态，特别是在Manga 109上的表现令人瞩目。