Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme (NeurIPS2021)

Overview

Prerequisites

• Linux
• Python 3
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Please type the command

  pip install -r requirements.txt

  to install dependencies.

Data preparation

• cityscapes
• horse2zebra
• celeb
• Coco, Set5, Set14, B100, Urban100

Pretrained Model

We provide a list of pre-trained models in link.

Pre-Training For Pruning

• Run the following script to pretrain a pix2pix on cityscapes dataset for generator pruning, all scripts for sagan, cyclegan, pix2pix, srgan can be found in ./scripts

  bash scripts/pix2pix/pretrain_for_pruning.sh

Training

• train lightweight generator using GCC

  bash scripts/pix2pix/train.sh

Testing

• test resulted models, FID or mIoU will be calculated, take pix2pix generator on cityscapes dataset as an example

  bash scripts/pix2pix/test.sh

Acknowledgements

Our code is developed based on DMAD, Self-Attention-GAN, pytorch-CycleGAN-and-pix2pix, a-PyTorch-Tutorial-to-Super-Resolution.