search

yan86471 / DMT-implementation

Implementation of Disentangled Makeup Transfer with Generative Adversarial Network
8 stars 2 forks source link
deep-learning disentangled-representations generative-adversarial-network makeup-transfer

readme

DMT

Introduction

A implementation of Disentangled Makeup Transfer with Generative Adversarial Network.

Training

  1. Download MT (Makeup Transfer) dataset from here.

  2. Put MT (Makeup Transfer) dataset to .\data\RawData. Your data path will like this:

    
.\data\RawData\images\makeup\*.png
.\data\RawData\images\non-makeup\*.png

.\data\RawData\segs\makeup*.png .\data\RawData\segs\non-makeup*.png



3. run `python convert.py`

4. Modify train.py and start training.  
    `python train.py` 

5. run `python export.py` and you will get h5 model in `.\Export`.

## Demo
1. make sure you have run `python export.py` to get h5 model.

2. Modify demo.py and run `python demo.py`, you will find the transfer result in `.\Transfer`.  

## Some issues to know
1. The test environment is
    - Python 3.7
    - tensorflow-gpu 2.0.0
    - tensorflow-addons 0.7.1
    - imgaug 0.4.0

2. This is still not a completed implementation, but almost 95% is the same as paper described.