Attribute2Sketch2Face

This repository is for our work on Face Synthesis from Visual Attributes via Sketch using Conditional VAEs and GANs.

Dependencies:

python 2.7

Pytorch

Data:

$root$ = attribute2sketch2face

(1) please download the preprocessed [CelebA] and LFWA database by link:

(2) extract them into $root$/dataset/

Training:

change the command direction into $root$
when training on the LFWA dataset. please run: python ./train_lfw.py
when training on the CelebA dataset. please run: python ./train_celeba.py

Testing:

when testingon the LFWA dataset. please run: python ./test_lfw.py
when testing on the CelebA dataset. please run: python ./test_celeba.py
All the test result should be in the directory: $root$/result/

Acknowledgement:

This code is highly inspired by the following:

cycleGAN and pix2pix:

StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks:

Generative Adversarial Text-to-Image Synthesis:

Also, we highly appreciate the help from He Zhang on the Densely Generator "G2" in this code.