# PSGAN-PyTorch

**Repository Path**: david_wu505/PSGAN-PyTorch

## Basic Information

- **Project Name**: PSGAN-PyTorch
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: GPL-3.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-02-11
- **Last Updated**: 2022-02-11

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# PSGAN-PyTorch

**PyTorch** implementation of [**PSGAN: Pose and Expression Robust Spatial-Aware GAN for Customizable Makeup Transfer**](https://arxiv.org/abs/1909.06956), still in construction...


## Preparation

- **Prerequisites**
    - PyTorch
    - Python 3.x with matplotlib, numpy and scipy

- **Dataset**
Dataset can be found in project page: http://colalab.org/projects/BeautyGAN

## Usage

Before training, you should generate train/test split labels using data_preparation/generate_labels.py. What you should do is just modify the data path in generate_labels.py.

The training setting is the same as [BeautyGAN](https://github.com/wtjiang98/BeautyGAN_pytorch). The implementation of PSGAN is still incomplete. I still have some problems in implementing the AMM with 68 landmarks detector.

However, the incomplete results is satisfying.

The training example of 50th epoch is as below:

![PSGAN_training_result](https://github.com/DateBro/PSGAN-PyTorch/blob/master/imgs/PSGAN_49_1259_fake.jpg)

The training example of BeautyGAN in 200th epoch is as below:

![BeautyGAN_training_result](https://github.com/DateBro/PSGAN-PyTorch/blob/master/imgs/BeautyGAN_199_1259_fake.jpg)

Though the implementation of PSGAN is still incomplete, it's obvious that PSGAN is pose and expression robust for makeup transfer.

## Acknowledgement

The code is built upon [BeautyGAN](https://github.com/wtjiang98/BeautyGAN_pytorch), thanks for their excellent work!