StyleGAN family is one of the most popular Generative Adversarial Networks (GANs) for unconditional generation. Despite its impressive performance, its high demand on storage and computation impedes their deployment on resource-constrained devices. This paper provides a comprehensive study of distilling from the popular StyleGAN-like architecture. Our key insight is that the main challenge of StyleGAN distillation lies in the output discrepancy issue, where the teacher and student model yield different outputs given the same input latent code. Standard knowledge distillation losses typically fail under this heterogeneous distillation scenario. We conduct thorough analysis about the reasons and effects of this discrepancy issue, and identify that the mapping network plays a vital role in determining semantic information of generated images. Based on this finding, we propose a novel initialization strategy for the student model, which can ensure the output consistency to the maximum extent. To further enhance the semantic consistency between the teacher and student model, we present a latent-direction-based distillation loss that preserves the semantic relations in latent space. Extensive experiments demonstrate the effectiveness of our approach in distilling StyleGAN2 and StyleGAN3, outperforming existing GAN distillation methods by a large margin.
翻译:StyleGAN 家庭是无条件一代人最受欢迎的创用Adversarial网络(GANs)之一。 尽管其业绩令人印象深刻,但对存储和计算的需求高,阻碍了在资源限制装置上的应用。本文对来自流行StyleGAN类似结构的蒸馏进行了全面研究。我们的主要见解是,StyleGAN 蒸馏的主要挑战在于产出差异问题,即教师和学生模型产生不同的产出,同时具有相同的输入潜在代码。标准知识蒸馏损失在这种混杂的蒸馏情景下通常会失败。我们对这一差异问题的原因和影响进行彻底分析,并确定绘图网络在确定生成图像的语义信息方面发挥着至关重要的作用。根据这一发现,我们为学生模型提出了一个新的初始化战略,可以最大限度地确保产出的一致性。为了进一步加强教师和学生模型之间的语义一致性,我们展示了一种基于潜在指令的蒸馏损失,从而保持了隐性空间的语义关系。我们进行了广泛的实验,展示了我们通过大变压GAN2 和SryGAN3 外演化GAN3 的方法在蒸馏Sle GranGAN 中的效率。