We use an existing implementation of deep convolutional GAN in pytorch to learn and generate chest X-Ray images directly. The dataset includes combined 5216 images of healthy people chest X-Rays and that of people with pneumonia. Later we also calculate the Inception score and FID score of the generated images.
It is only a proof-of-concept exercise and is in no way fine-tuned.