Conditional Adversarial Networks (CAN)

This is an tensorflow implementation of pix2pix based on 'image-to-image translation with conditional adversarial networks'.

Requirements

• Nvidia GPU and cuDNN 5.0
• Tensorflow 1.0.0

Train

• Clone this repository

git clone https://github.com/fullfanta/pix2pix-tensorflow.git

• Download database I borrowed download script from pix2pix.

$ cd pix2pix-tensorflow
$ sh datasets/download_datasets.sh facades
$ sh datasets/download_datasets.sh edges2handbags
$ sh datasets/download_datasets.sh edges2shoes
$ sh datasets/download_datasets.sh maps

• Train CAN with specific dataset.

$ python train.py --dataset=facades --which_direction=BtoA --batch_size=1 --device=gpu

I attatch two shell script to train facades(train_facades.sh) and edges2handbags(train_edges2handbags.sh).

• To see training procedure, you can start tensorboard.

$ tensorboard --logdir=result_facades

Results

These are some results of facades: left - input, center - gt and right - generated image.

By modifying this code, I transformed facial expression from neutral to smile by using MultiPIE.