lofgan-pytorch's Introduction

LoFGAN-pytorch

The official pytorch implementation of our paper LoFGAN: Fusing Local Representations for Few-shot Image Generation, ICCV 2021.

LoFGAN: Fusing Local Representations for Few-shot Image Generation

Zheng Gu, Wenbin Li, Jing Huo, Lei Wang, and Yang Gao

Paper

Prerequisites

Pytorch 1.5

Preparing Dataset

Download the datasets and unzip them in datasets folder.

Training

python train.py --conf configs/flower_lofgan.yaml \
--output_dir results/flower_lofgan \
--gpu 0

You may also customize the parameters in configs.
It takes about 30 hours to train the network on a V100 GPU.

Testing

python test.py --name results/flower_lofgan --gpu 0

The generated images will be saved in results/flower_lofgan/test.

Evaluation

python main_metric.py --gpu 0 --dataset flower \
--name results/flower_lofgan \
--real_dir datasets/for_fid/flower --ckpt gen_00100000.pt \
--fake_dir test_for_fid

Citation

If you use this code for your research, please cite our paper.

@inproceedings{gu2021lofgan,
title={LoFGAN: Fusing Local Representations for Few-Shot Image Generation},
author={Gu, Zheng and Li, Wenbin and Huo, Jing and Wang, Lei and Gao, Yang},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={8463--8471},
year={2021}
}