🚀 Drishti - A Pix2Pix GAN Model

📝 Introduction

This project aims to develop an advanced system that uses pix2pix GAN (Generative Adversarial Network). This model will address challenges in understanding SAR (Synthetic Aperture RADAR). It converts SAR images captured by satellite to optical images, making it easier for non-experts to interpret and analyze satellite data. This approach can be valuable for applications in remote sensing, disaster monitoring, and environmental analysis.

---

🎯 Goals

• SAR-to-Optical Translation – Convert SAR images into optical images.

• High-Quality Image Generation – Generate realistic and coherent optical images using Pix2Pix GAN.

• User-Friendly Interface – Provide an intuitive Streamlit interface to upload SAR images and view generated results.

---

✨ Features

• Upload SAR images and generate optical images instantly.
• High-quality image translation with Pix2Pix GAN.
• Easy-to-use interface using Streamlit.
• Supports deployment on standard CPU/GPU hardware.

---

🏗️ Architecture

1. Generator

2. Discriminator

3. Generative Adversarial Network

---

⚡Loss Functions Used

Component	Loss Function Used	Purpose
Generator (G)	L1 Loss + Adversarial Loss (BCE)	Encourages generator to produce realistic optical images close to target.
Discriminator (D)	Binary Cross-Entropy (BCE) Loss	Helps discriminator distinguish real vs fake images.

Model Link : https://huggingface.co/roshan-acharya/dristhi-generator

📂 Experiment Details

📂Datasets

Dataset	Use	Samples
SAR-Optical Pairs (Custom)	Training	15,000+
SAR-Optical Pairs (Validation)	Validation	2,500+

🛠️ Training Configuration

Parameter	Value
batch_size	16
grad_clip	0.5
learning_rate (Adam)	2e-4
scheduler	StepLR
epochs	100
image_size	256×256

💻 Installation

1. Clone the repository:

   git clone https://github.com/roshan-acharya/Drishti
   cd Drishti

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Streamlit application:

   streamlit run app.py

---

👥 Collaborators

Roshan Acharya

Loblesh Bhartal

---

📚 References / Citations

For further reference and proper attribution, please cite:

1. Pix2Pix GAN (Paper)
   Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, Alexei A. Efros.
   Image-to-Image Translation with Conditional Adversarial Networks, CVPR 2017.
   PDF

2. Medium Article
   Pix2Pix GAN for Generating Map Given Satellite Images Using PyTorch
   Medium Article Link