This is a Pytorch Implementation for learning an image-to-image translation. The original paper is called Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks. Instead of using default generator network, we use two other network namely DnCNN and UNet. Detail structure can be found in this paper. Some of our result is shown below:
Custom python module: image_tool.py, nntools.py
model.py ---The net which is just the model downloaded from Internet
DnCNNmodel.py ---The net with DnCNN as generator
Unetmodel.py ---The net with Unet as generator
dataset.py ---The .py file help to load the image
image_pool.py ---The .py file which buffer the images(downloaded from Internet)
train_CycleGAN.ipynb ---The .ipynb file which run the experiments to obtain model
Demo_CycleGAN.ipynb ---Use the trained model to do testing, 'the trained model is stored in output2
MyCycleGan-DnCNN.ipynb ---Training record with DnCNN as generator
MyCycleGan-Unet.ipynb ---Training record with Unet as generator