Distilling portable Generative Adversarial Networks for Image Translation
This work addresses the challenge of making GANs practical for mobile applications, representing an incremental improvement in network compression for generation tasks.
The paper tackles the problem of deploying Generative Adversarial Networks (GANs) for image translation on mobile devices by reducing their computational and storage costs, achieving strong performance through a knowledge distillation approach with a student generator and discriminator.
Despite Generative Adversarial Networks (GANs) have been widely used in various image-to-image translation tasks, they can be hardly applied on mobile devices due to their heavy computation and storage cost. Traditional network compression methods focus on visually recognition tasks, but never deal with generation tasks. Inspired by knowledge distillation, a student generator of fewer parameters is trained by inheriting the low-level and high-level information from the original heavy teacher generator. To promote the capability of student generator, we include a student discriminator to measure the distances between real images, and images generated by student and teacher generators. An adversarial learning process is therefore established to optimize student generator and student discriminator. Qualitative and quantitative analysis by conducting experiments on benchmark datasets demonstrate that the proposed method can learn portable generative models with strong performance.