QGAN: Quantized Generative Adversarial Networks

TL;DR

This paper introduces QGAN, a novel quantization method for GANs that reduces model size to 1-2 bits while maintaining high sample quality, enabling deployment on edge devices.

Contribution

The paper presents a new EM-based quantization approach for GANs and a multi-precision algorithm to optimize bit-width for quality and efficiency.

Findings

QGAN can quantize GANs to 1-2 bits.

Quantized GANs achieve comparable quality to original models.

The method is effective on CIFAR-10 and CelebA datasets.

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 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 methods to quantize GANs. Our observation reveals that none of them generates samples with reasonable quality because of the underrepresentation of quantized values in model weights, and the generator and discriminator networks show different sensitivities upon quantization methods. 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 the optimal number of bits of quantized GAN models in conjunction with corresponding result qualities. Experiments on CIFAR-10 and CelebA show that QGAN can quantize GANs to even 1-bit or 2-bit representations with results of quality comparable to original models.