Check lists for Cycle GAN
We have come a long way from where we’ve started. Lets see what we have done till now and also watch out for upcoming stuff!
- Understanding & Implementing - Convolutional Neural Networks.
DONE - Understanding & Implementing - Simplified Generative Adversarial Networks.
DONE - Theory behind cycle consistent image 2 image translation with GANs.
DONE - Implementing Cycle GAN from scratch.
DONE - Analyzing different datasets with our network.
It’s time to test our implementation on slandered datasets and analyze the performance of the network.
Implementation
If you want to implement our code off the shelf, you can find the entire code for Cycle GAN network in our repository. To download specific datasets, please refer to the Resources section of the About Page.
Dependencies
Note : The following packages must be installed in your machine if you want to run CycleGAN :
- glob
- Tensorflow
- scipy
- numpy
- Pillow
The dependencies can be installed by following the commands :
git clone https://github.com/CycleGANS/CS543CycleGANsProject.git
sh ./download_dataset.sh horse2zebra
mv datasets Datasets
pip install -r requirements.txt
Running the code
Note : This is a heavy code to execute on a CPU. A GPU is highly recommended. We have used Blue Waters - K80 Graphical Processing Unit for this project. You can also download our trained weights available here and keep it in the checkpoints directory.
You can run our code by following command:
python3 main.py
Results
We have tested our network implementation in 5 different datasets : horse2zebra, apple2orange, Summer2winter, monet2photo and vangogh2photo . Here are some of the results on each of the following :
Horse - Zebra
Summer - Winter
Monet - Photo
