Wavelet Knowledge Distillation: Towards Efficient Image-to-Image Translation

TL;DR

This paper introduces wavelet knowledge distillation for GANs, focusing on high frequency information to improve efficiency and reduce model size without sacrificing image quality in image-to-image translation.

Contribution

It proposes a novel wavelet-based knowledge distillation method that emphasizes high frequency bands, enabling significant model compression and acceleration with minimal performance loss.

Findings

Achieved 7.08x model compression and 6.80x speedup on CycleGAN.

Focused distillation on high frequency bands improves image quality.

Discriminator compression enhances generator performance.

Abstract

Remarkable achievements have been attained with Generative Adversarial Networks (GANs) in image-to-image translation. However, due to a tremendous amount of parameters, state-of-the-art GANs usually suffer from low efficiency and bulky memory usage. To tackle this challenge, firstly, this paper investigates GANs performance from a frequency perspective. The results show that GANs, especially small GANs lack the ability to generate high-quality high frequency information. To address this problem, we propose a novel knowledge distillation method referred to as wavelet knowledge distillation. Instead of directly distilling the generated images of teachers, wavelet knowledge distillation first decomposes the images into different frequency bands with discrete wavelet transformation and then only distills the high frequency bands. As a result, the student GAN can pay more attention to its learning on high frequency bands. Experiments demonstrate that our method leads to 7.08 times compression and 6.80 times acceleration on CycleGAN with almost no performance drop. Additionally, we have studied the relation between discriminators and generators which shows that the compression of discriminators can promote the performance of compressed generators.