发表机构
North South University（北南大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出GAN-Diff框架，将预训练WGAN-GP特征作为先验耦合到条件扩散U-Net，在CelebA数据集的高斯去噪和2倍超分辨率任务上，分别提升PSNR达4.40 dB和3.70 dB，实现稳定高效的图像复原。
AI 中文摘要
生成对抗网络（GAN）可实现高效图像生成，而扩散模型能提供高质量图像复原，但需迭代采样。本文提出一种混合GAN引导的扩散框架，使用预训练带梯度惩罚的Wasserstein GAN（WGAN-GP）作为条件扩散图像复原的特征先验。冻结的WGAN-GP生成器的中间特征通过交叉注意力被整合到扩散U-Net中，且在DDIM采样过程中保持固定。该框架在两个复原任务（高斯去噪和2倍超分辨率）上使用CelebA人脸图像进行评估。开发过程中，识别并解决了多种不稳定性来源，包括对抗学习率失衡、不当的扩散初始化、过度损坏以及参数平均不足。最终框架持续提升退化图像和低分辨率图像的质量，与各自输入基线相比，去噪性能的峰值信噪比（PSNR）提升了4.40 dB，超分辨率性能提升了3.70 dB。这些结果表明，冻结的GAN特征先验具有引导扩散模型实现稳定且有效图像复原的潜力。
英文摘要
Generative adversarial networks (GANs) can provide efficient image generation, while diffusion models offer high-quality image restoration but require iterative sampling. This paper presents a hybrid GAN-guided diffusion framework that uses a pretrained Wasserstein GAN with gradient penalty (WGAN-GP) as a feature prior for conditional diffusion-based image restoration. Intermediate features from the frozen WGAN-GP generator are incorporated into a diffusion U-Net through cross-attention and remain fixed during the DDIM sampling process. The framework is evaluated on two restoration tasks, Gaussian denoising and 2Xsuper-resolution, using CelebA face images. During development, several sources of instability were identified and addressed, including adversarial learning-rate imbalance, inappropriate diffusion initialization, excessive corruption, and insufficient parameter averaging. The resulting framework consistently improves the quality of both degraded and low-resolution images. In particular, it improves denoising performance by 4.40 dB in PSNR and super-resolution performance by 3.70 dB over their respective input baselines. These results demonstrate the potential of a frozen GAN feature prior to guide diffusion models toward stable and effective image restoration.
Comments7 pages, 10 figures, 4 tables
