databill86 / uaggan

A Pytorch implementation of "Unsupervised Attention-Guided Image-to-Image Translation"

uaggan

A Pytorch implementation of "Unsupervised Attention-Guided Image-to-Image Translation", NIPS 2018, [Paper] | [TF code]

---

Installation

For Conda users, we provide a installation script ./scripts/conda_deps.sh. Alternatively, you can create a new Conda environment using this command:

conda env create -f environment.yml

Prepare Dataset:

• Download a CycleGAN dataset (e.g. horse2zebra, apple2orange):

bash ./datasets/download_cyclegan_dataset.sh horse2zebra

• Use your own dataset: put images from each domain at folder_a and folder_b respectively.

Usage

• Training

sh ./scripts/train_uaggan.sh <gpu_id>

or

nohup sh -u ./scripts/train_uaggan.sh <gpu_id> > uag.log &

• Test

sh ./scripts/test_uaggan.sh <gpu_id>

Mapping results

• horse2zebra (without early stopping, threshold=0.1)

Real A	Attention Mask	Fake B	Real B	Attention Mask	Fake A

[More Results >>>]

• horse2zebra (with early stopping, threshold=0.1)

//TODO

• apple2orange

//TODO

Faster version

Combining the attention module and domain translation module into one generator, inspired by GANimation. (Results are not good.)

• Training

sh ./scripts/train_faster_uaggan.sh <gpu_id>

• Testing

sh ./scripts/test_faster_uaggan.sh <gpu_id>

Real A	Attention Mask	Fake B	Real B	Attention Mask	Fake A

---

Acknowledgment

This code is based on the pytorch-CycleGAN-and-pix2pix. Thanks to the contributors of this project.