Terra_Mask 🛰️

Land Cover Semantic Segmentation using PyTorch

---

📌 What Is This Project?

Terra_Mask is an end-to-end computer vision project for semantic segmentation in land cover classification. It uses deep learning with PyTorch to classify features in high-resolution satellite imagery — such as buildings, woodland, water, and roads.

Built primarily for the LandCover.ai dataset, this project is modular and can be adapted to any semantic segmentation dataset.

---

🌍 What Does It Do?

This project enables you to:

• ✅ Train segmentation models on land cover images
• ✅ Test and infer with trained models
• ✅ Prompt specific classes for prediction without retraining

Promptable Prediction: After training the model on all classes, you can selectively visualize only specific classes (e.g., just "buildings" and "water") by modifying the test_classes in the config.

---

📷 Screenshots

🧠 How It Works

1. Data Preparation

• Patches are created from large aerial images.
• Background-heavy patches are discarded.
• Data is split into train/val/test sets.

2. Model Architecture

• Uses UNet from segmentation_models_pytorch
• EfficientNet as the encoder backbone
• Decoder upsamples features to produce segmentation masks

3. Training

• Uses Dice Loss for segmentation accuracy
• Learning rate scheduling for better optimization

4. Inference

• Patches are passed through the model
• Reconstructed masks are filtered by class
• Visualizations are generated and saved

---

🚀 Quick Start

📁 Dataset Setup

Copy the dataset from LandCover.ai on Kaggle

Organize it as:

data/
├── train/
│ ├── images/ # Training satellite images
│ └── masks/ # Corresponding segmentation masks
└── test/
├── images/ # Testing satellite images
└── masks/ # Corresponding segmentation masks

🔧 Environment Setup

# Clone this repository
git clone https://github.com/souvikmajumder26/Land-Cover-Semantic-Segmentation-PyTorch.git
cd Land-Cover-Semantic-Segmentation-PyTorch

# Install required packages
pip install -r requirements.txt

🏃‍♂️ How to Run

🔬 Testing with Pre-trained Model

cd src
python test.py

🎯 Inference

cd src
python inference.py

🏋️‍♂️ Training the Model(⚠️ Requires significant GPU resources.)

cd src
python train.py

🐳 Docker Support

Build and Run in Docker

docker build -t segment_project_image .
docker run --name segment_container -d segment_project_image

Copy Outputs

docker cp segment_container:/segment_project/models ./models
docker cp segment_container:/segment_project/logs ./logs
docker cp segment_container:/segment_project/output ./output

Cleanup

docker stop segment_container
docker rm segment_container
docker rmi segment_project_image

📁 Project Structure

├── config/
│   └── config.yaml        # Configuration file (classes, model params)
├── data/
│   ├── train/
│   ├── test/
├── models/                # Pretrained weights
├── output/
│   ├── predicted_masks/   # .tif mask outputs
│   ├── prediction_plots/  # .png visualizations
├── logs/                  # Training and testing logs
├── src/
│   ├── train.py
│   ├── test.py
│   └── inference.py
├── requirements.txt
└── README.md

🔧 Technologies Used

🧠 Deep Learning & Vision

• PyTorch 2.0.1
• torchvision 0.15.2
• segmentation_models_pytorch 0.3.3
• EfficientNet (efficientnet-pytorch)
• timm (PyTorch Image Models)

🧰 Image & Data Handling

• OpenCV (opencv-python-headless)
• patchify
• scikit-image, scikit-learn
• albumentations
• tifffile, Pillow, NumPy

📦 Utilities

• PyYAML
• tqdm
• split-folders
• filelock, joblib
• pretrainedmodels
• huggingface-hub

📈 Loss and Optimization

• Dice Loss
• Adam Optimizer
• LR Scheduler (PyTorch)

🐳 Deployment

• Docker

📜 License

This project is released under the MIT License.