Dynamically Grown Generative Adversarial Networks
This work addresses the need for automated architecture design in GANs for image generation, offering an incremental improvement over existing progressive growing methods.
The paper tackles the problem of manually designing progressive growing strategies for large GANs by proposing a method to dynamically grow GANs during training, optimizing both architecture and parameters automatically, which achieves new state-of-the-art results in image generation.
Recent work introduced progressive network growing as a promising way to ease the training for large GANs, but the model design and architecture-growing strategy still remain under-explored and needs manual design for different image data. In this paper, we propose a method to dynamically grow a GAN during training, optimizing the network architecture and its parameters together with automation. The method embeds architecture search techniques as an interleaving step with gradient-based training to periodically seek the optimal architecture-growing strategy for the generator and discriminator. It enjoys the benefits of both eased training because of progressive growing and improved performance because of broader architecture design space. Experimental results demonstrate new state-of-the-art of image generation. Observations in the search procedure also provide constructive insights into the GAN model design such as generator-discriminator balance and convolutional layer choices.