QGAN: Quantize Generative Adversarial Networks to Extreme low-bits

Peiqi Wang, Yu Ji, Xinfeng Xie, Yongqiang Lyu, Dongsheng Wang, Yuan Xie

Keywords: generative adversarial networks, quantization, extreme low bits

Abstract: The intensive computation and memory requirements of generative adversarial neural networks (GANs) hinder its real-world deployment on edge devices such as smartphones. Despite the success in model reduction of convolutional neural networks (CNNs), neural network quantization methods have not yet been studied on GANs, which are mainly faced with the issues of both the effectiveness of quantization algorithms and the instability of training GAN models. In this paper, we start with an extensive study on applying existing successful CNN quantization methods to quantize GAN models to extreme low bits. Our observation reveals that none of them generates samples with reasonable quality because of the underrepresentation of quantized weights in models, and the generator and discriminator networks show different sensitivities upon the quantization precision. Motivated by these observations, we develop a novel quantization method for GANs based on EM algorithms, named as QGAN. We also propose a multi-precision algorithm to help find an appropriate quantization precision of GANs given image qualities requirements. Experiments on CIFAR-10 and CelebA show that QGAN can quantize weights in GANs to even 1-bit or 2-bit representations with results of quality comparable to original models.