Coarse-to-Fine Searching for Efficient Generative Adversarial Networks
This work addresses the challenge of designing efficient GANs for image generation, which is incremental as it builds on existing NAS methods to reduce search costs and improve performance.
The paper tackles the problem of neural architecture search for efficient generator networks in GANs by proposing a coarse-to-fine search strategy and a fair supernet training approach, resulting in models with better image quality and lower computational costs, such as a 2.56 MB model with a 24.13 FID score in 8 GPU hours on edges-to-shoes and a 1.49 MB model with a 24.94 PSNR score in 10 GPU hours on Urban100.
This paper studies the neural architecture search (NAS) problem for developing efficient generator networks. Compared with deep models for visual recognition tasks, generative adversarial network (GAN) are usually designed to conduct various complex image generation. We first discover an intact search space of generator networks including three dimensionalities, i.e., path, operator, channel for fully excavating the network performance. To reduce the huge search cost, we explore a coarse-to-fine search strategy which divides the overall search process into three sub-optimization problems accordingly. In addition, a fair supernet training approach is utilized to ensure that all sub-networks can be updated fairly and stably. Experiments results on benchmarks show that we can provide generator networks with better image quality and lower computational costs over the state-of-the-art methods. For example, with our method, it takes only about 8 GPU hours on the entire edges-to-shoes dataset to get a 2.56 MB model with a 24.13 FID score and 10 GPU hours on the entire Urban100 dataset to get a 1.49 MB model with a 24.94 PSNR score.