carime-pytorch's Introduction

CariMe-pytorch

The official pytorch implementation of our TMM paper CariMe: Unpaired Caricature Generation with Multiple Exaggerations.

CariMe: Unpaired Caricature Generation with Multiple Exaggerations

Zheng Gu, Chuanqi Dong, Jing Huo, Wenbin Li, and Yang Gao

Paper: https://ieeexplore.ieee.org/abstract/document/9454341/

Prerequisites

Python 3.6
Pytorch 1.5.1
scikit-image 0.17.2

Preparing Dataset

Get the Webcaricature dataset, unzip the dataset to the data folder and align the dataset by running the following script:

python alignment.py

Training

Train the Warper:

python train_warper.py

Train the Styler:

python train_styler.py

Testing

Test the Warper only:

python test_warper.py --scale 1.0

Test the Styler only:

python test_styler.py

Generate caricatures with both exaggeration and style transfer:

python main_generate.py --model_path_warper pretrained/warper.pt --model_path_styler pretrained/styler.pt

Generate caricatures with both exaggeration and style transfer for a single image:

python main_generate_single_image.py --model_path_warper pretrained/warper.pt --model_path_styler pretrained/styler.pt --input_path images/Meg Ryan/P00015.jpg --generate_num 5 --scale 1.0

The above command will translate the input photo into 5 caricatures with different exaggerations and styles:

Pretrained Models

The pre-trained models are shared here.

Citation

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

@article{gu2021carime,
title={CariMe: Unpaired Caricature Generation with Multiple Exaggerations},
author={Gu, Zheng and Dong, Chuanqi and Huo, Jing and Li, Wenbin and Gao, Yang},
journal={IEEE Transactions on Multimedia},
year={2021},
publisher={IEEE}
}