README.md

Team Pandas Submission for the Problem Statement 1 - Neural Style Transfer at Satellite Imagery Track Shaastra 2023 Indian Institute of Technology (IIT), Madras .

SAR-to-RGB-CycleGAN 🎆

This repository contains the implementation of a Cycle Generative Adversarial Network (CycleGAN) model for converting Synthetic Aperture Radar (SAR) satellite imagery to Photorealistic-RGB (PS-RGB) optical images.

Downloading the dataset: 💽

Prerequisites:

• AWS Account (create one if you dont have one)
• Install awscli (Run this on bash/terminal depending on your OS)

pip install awscli

Procedure 🚲

• Create a data folder, inside which,Cceate 2 new directories SAR-Intensity & PS-RGB(optical images)

In SAR-Intensity folder execute:

aws s3 sync s3://spacenet-dataset/spacenet/SN6_buildings/train/AOI_11_Rotterdam . --exclude "*" --include "SAR-Intensity/*"

In PS-RGB folder execute:

aws s3 sync s3://spacenet-dataset/spacenet/SN6_buildings/train/AOI_11_Rotterdam . --exclude "*" --include "PS-RGB/*"

Presentation: 🎬

Link

Notebook Links 📙

Data Preprocessing and Model Training

• Notebook

• Colab

Generating Optical Images with Trained Model

• Colab

Other resources on drive 🚗

• Model

Generated Samples (5 images as mentioned) 📸

---

Made with ❤️ and 💻