Face-Parsing-Network

PyTorch implementation of Face Parsing (based on semantic segmentation)

Network originally based on : Object contour detection with a fully convolutional encoder-decoder network [J. Yang, 2016]

Used for : Generative Face Completion [Y. Li, 2017]

Used Dataset : CelebA-HQ Dataset

The network layer codes(like batch normalization, convolution...) was refer to Context Attention

• Dataset and ground-truth should be located as:
  • Training image directory
    • 1.jpg
    • 2.jpg
  • Ground-truth image directory
    • 1.png
    • 2.png

• Trained PyTorch model from CelebA-HQ dataset : Here
• Samples of training and ground-truth set : Download

Quick inference using pre-trained model

$ cp parser_00100000.pt checkpoints/
$ python seg_inference.py

Then samples are saved in output directory

Trained with 30,000 CelebA-HQ dataset and applied it to LFW and CelebA dataset using that pre-trained model

Python 3.6 + Pytorch 1.1