PhotoGAN

Generation of Photo Realistic Image using GAN and SPADE. This project demonstrates the process of Image-to-Image Translation in which image is transformed to another form of image.

Here our Goal is to Generate Photo Realistic Image, using the SPADE Architecture. This Architecture helps to Generate more Robust Images using GAN.

Setup

1. In-order to setup we require python 3.x version (Better with Anaconda Environment) with requirements specified in requirements.txt.
2. Download the checkpoints from this link.
3. Extract the checkpoints in the current directory ie. PhotoGAN > Checkpoints.

4. To test the output from the model some data are already provided, to test those write the following command in the conda environment

   python test.py --name coco_pretrained --dataset_mode coco --dataroot datasets/coco_stuff/

Visuals

Image to Image Synthesis is performed over the Drawing (Left) and the Realistic Image is Generated (Right).

Contributors

Special Thanks to project partners and

1. Akshaya M
2. Shubham Dogra
3. Adarsh Revankar

and also

Big thanks to Synchronized BatchNormalization for providing Batch Normalization implementation.