VaryingSkinTone

This repository contains the Keras implementation of the paper titled
"An Unsupervised Approach towards Varying Human Skin Tone Using Generative Adversarial Networks" https://ieeexplore.ieee.org/abstract/document/9412852
Arxiv: https://arxiv.org/abs/2010.16092

This is an inference code. The checkpoints and a sample dataset are upload at, https://drive.google.com/drive/folders/1eVO9ki1drp1fGjrucNgd24iL1A6MB_0Y?usp=sharing

Testing:

This code can be run in 2 ways:

1. Using our trained segmentation model the skin segmentation part To execute in this case:
   set the following in the params.py file
   dataset_path = './Datasets/DeepFashion/Category-and-Attribute-Prediction-Benchmark/'
   Command to run:
   python2 stage/test_skin.py -batch_size 1 -range_count 10 -seg_choice False -test_filename deepfashion_names_upper.txt
   

2. Using pre-computed segmentations by some other method of your choice and then running our skin tone changing model on the provided image and the estimated segmentation.
   To execute in this case:
   set the following in the params.py file
   dataset_path = './Datasets/MyData/'
   Command to run:
   python2 stage/test_skin.py -batch_size 1 -range_count 10 -seg_choice True -test_filename test_external.txt
   

---

The code is tested in the following versions:
keras = '2.2.4' tensorflow = '1.14.0' python = '2.7.12'

In case you use this code please consider citing

  @inproceedings{roy2021unsupervised,
  title={An Unsupervised Approach towards Varying Human Skin Tone Using Generative Adversarial Networks},
  author={Roy, Debapriya and Mukherjee, Diganta and Chanda, Bhabatosh},
  booktitle={2020 25th International Conference on Pattern Recognition (ICPR)},
  pages={10681--10688},
  year={2021},
  organization={IEEE}}