Vis5 / face2passport

StyleGAN2 - Official TensorFlow Implementation

Introduction

This repo was cloned & edited from https://github.com/NVlabs/stylegan2.

This project aims to generate formal passport photo from informal photos with human face. It includes 500_128_passport dataset which consists of over 500 passport photos of size 128x128 ready to train.

Requirements

This repo was tested on Ubuntu 18.04 and Tensorflow 1.14 with GPU support enabled.

pip3 install -r requirements.txt

Training

Already trained model is in the results folder. If it is not present, please download from https://drive.google.com/drive/folders/1UbUAbGyZInt6Qt4PDI2s6YR8SAfexuUh

To generate your own model:

python3 run_training.py --num-gpus=1 --data-dir=datasets --config=config-f --dataset=500_128_passport

Encoding

Put images you want to convert into raw_images/ dir and run the following command. Generated results will be in the generated_images/ directory.

python3 encode.py

and then

python3 translate.py