This repo is the pytorch implementation of pix2pix architecture as provided in the paper Image-to-Image Translation with Conditional Adversarial Networks. Only maps Dataset has been used for training and testing.
How to use?
- Clone the repo
- Download the Maps dataset and unzip it into cloned repo (working directory). This will create a directory called “maps”, with "train" and "val" folderinside it.
- To train, run train.py
Each image in the dataset is in JPEG format with 1200px * 600px dimensions, containing satellite image with its correspondng map. A sample image has been shown below.
Sample after splitting the raw image into seperate "Satellite" and corresponding "map" images.
Pix2pix architecture uses Unet architecture as a Generator and PatchGAN as a Discriminator.
patchGAN architecture used as Discriminator as shown below.
