Home

Awesome

License CC BY-NC-SA 4.0 Python 3.7

FUNIT: Few-Shot Unsupervised Image-to-Image Translation

animal swap gif

Project page | Paper | FUNIT Explained | GANimal Demo Video | Have fun with GANimal

Few-shot Unsueprvised Image-to-Image Translation<br> Ming-Yu Liu, Xun Huang, Arun Mallya, Tero Karras, Timo Aila, Jaakko Lehtinen, and Jan Kautz.<br> In arXiv 2019.

License

Copyright (C) 2019 NVIDIA Corporation.

All rights reserved. Licensed under the CC BY-NC-SA 4.0 (Attribution-NonCommercial-ShareAlike 4.0 International)

The code is released for academic research use only. For commercial use, please contact researchinquiries@nvidia.com.

For press and other inquiries, please contact Hector Marinez

Installation

To reproduce the results reported in the paper, you would need an NVIDIA DGX1 machine with 8 V100 GPUs.

Hardware Requirement

To reproduce the experiment results reported in our ICCV paper, you would need to have an NVIDIA DGX1 machine with 8 V100 32GB GPUs. The training will use all 8 GPUS and take almost all of the GPU memory. It would take about 2 weeks to finish the training.

Dataset Preparation

Animal Face Dataset

We are releasing the Animal Face dataset. If you use this dataset in your publication, please cite the FUNIT paper.

cd dataset
wget http://www.image-net.org/challenges/LSVRC/2012/nnoupb/ILSVRC2012_img_train.tar
tar xvf ILSVRC2012_img_train.tar
python tools/extract_animalfaces.py datasets/ILSVRC/Data/CLS-LOC/train --output_folder datasets/animals --coor_file datasets/animalface_coordinates.txt

Training

Once the animal face dataset is prepared, you can train an animal face translation model by running.

python train.py --config configs/funit_animals.yaml --multigpus

The training results including the checkpoints and intermediate results will be stored in outputs/funit_animals.

For custom dataset, you would need to write an new configuration file. Please create one based on the example config file.

Testing pretrained model

To test the pretrained model, please first create a folder pretrained under the root folder. Then, we need to downlowad the pretrained models via the link and save it in pretrained. Untar the file tar xvf pretrained.tar.gz.

Now, we can test the translation

python test_k_shot.py --config configs/funit_animals.yaml --ckpt pretrained/animal149_gen.pt --input images/input_content.jpg --class_image_folder images/n02138411 --output images/output.jpg

The above command with translate the input image

images/input_content.jpg

input image

to an output meerkat image

output image

by using a set of 5 example meerkat images

Citation

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

@inproceedings{liu2019few,
  title={Few-shot Unsueprvised Image-to-Image Translation},
  author={Ming-Yu Liu and Xun Huang and Arun Mallya and Tero Karras and Timo Aila and Jaakko Lehtinen and Jan Kautz.},
  booktitle={arxiv},
  year={2019}
}