This is a Tensorflow implementation of Image-to-Image Translation with Conditional Adversarial Networks that aims to infer a mapping from X to Y, where X is a single channel "black and white" image and Y is 3-channel "colorized" version of that image.

We make use of Generative Adversarial Networks conditioned on the input to teach a generative neural network how to produce our desired results.

The purpose of this repository is to port the model over to TensorFlow.

Input	Output	Ground-Truth

• Python 3.5
• TensorFlow (>= r0.12)

python train.py

python test.py 'path/to/input/image'

• colorize.py

  Main script that interprets the user's desired actions through parsed arguments.

• generator.py

  Contains the generative net that can colorize single-channel images when trained.

• discriminator.py

  Contains the discriminative net that can discriminate between synthesized colorized images and ground-truth images.

• net.py

  Contains the neural network super class with universal layer construction and instance normalization methods.

• train.py

  Contains a Trainer class that is responsible for training the generative adversarial networks and any related routines such as retrieving training data.