用无条件判别器提升GAN训练稳定性，避免模式崩溃。
UCD: Unconditional Discriminator Promotes Nash Equilibrium in GANs
- 让判别器不接受条件输入，强制学习更全面特征
- 在ImageNet-64上达1.47 FID，优于StyleGAN-XL等模型
- 可即插即用，适合需要稳定生成的科研与工程场景
对抗训练是单步生成的关键，尤其在生成对抗网络（GAN）和扩散模型蒸馏中。然而实际中GAN训练常难收敛且易发生模式崩溃。本文定量分析了GAN训练中纳什均衡的程度，发现向判别器输入条件会造成冗余捷径，阻碍有意义知识提取。为此提出无条件判别器（UCD），强制判别器在无条件输入下提取更全面、鲁棒的特征，从而以更好知识监督生成器，促进纳什均衡。理论证明UCD与原始GAN框架兼容，可直接嵌入。大量实验验证其高效显著提升性能：在ImageNet-64上达到1.47 FID，优于StyleGAN-XL及多个先进单步扩散模型。代码将公开。
原文摘要 · Abstract (English)
Adversarial training turns out to be the key to one-step generation, especially for Generative Adversarial Network (GAN) and diffusion model distillation. Yet in practice, GAN training hardly converges properly and struggles in mode collapse. In this work, we quantitatively analyze the extent of Nash equilibrium in GAN training, and conclude that redundant shortcuts by inputting condition in $D$ disables meaningful knowledge extraction. We thereby propose to employ an unconditional discriminator (UCD), in which $D$ is enforced to extract more comprehensive and robust features with no condition injection. In this way, $D$ is able to leverage better knowledge to supervise $G$, which promotes Nash equilibrium in GAN literature. Theoretical guarantee on compatibility with vanilla GAN theory indicates that UCD can be implemented in a plug-in manner. Extensive experiments confirm the significant performance improvements with high efficiency. For instance, we achieved \textbf{1.47 FID} on the ImageNet-64 dataset, surpassing StyleGAN-XL and several state-of-the-art one-step diffusion models. The code will be made publicly available.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。