Image Segmentation Using U-Net

This project focuses on image segmentation using the U-Net architecture. The goal of the project is to train a U-Net model to perform pixel-level classification on a dataset, specifically for segmentation tasks. The project is divided into several stages including data preprocessing, model training, and evaluation.

Repository Structure

The repository contains the following key files and scripts:

1. image_segmentation.ipynb

• A Jupyter notebook that demonstrates the process of image segmentation using the U-Net model. This includes loading the dataset, preprocessing, loading custom model weights and evaluating the results.

2. model_training.ipynb

• A Jupyter notebook dedicated to training the U-Net model on the processed dataset. It covers the training loop, loss functions, optimizers and evaluation metrics used to measure the performance of the model.

3. UNet.py

• A Python script that defines the U-Net model architecture. This script includes the U-Net implementation with customizable layers and configurations for image segmentation tasks.

4. data_prep.py

• A Python script that handles data preprocessing. This script processes raw data, including image resizing, normalization of mask and splitting dataset, to prepare it for model training.

Key Features

• U-Net Architecture: Implements a robust U-Net model, which is widely used for biomedical image segmentation.
• Data Preprocessing: Ensures that the data is appropriately prepared for training with resizing and normalization.
• Model Training: Supports training with different hyperparameters such as batch size, learning rate, and early stopping.
• Evaluation Metrics: Uses metrics like Intersection over Union (IoU) and Dice coefficient to evaluate the performance of the model.

Requirements

To run the project, you'll need the following Python packages:

pip install tensorflow keras numpy pandas matplotlib seaborn scikit-learn opencv-python

How to Run

1. Model Training: Use model_training.ipynb to train the U-Net model on the prepared dataset.
2. Segmentation: Perform image segmentation using image_segmentation.ipynb, which includes evaluation on test data.
3. Custom Scripts: You can modify UNet.py for model architecture adjustments and use data_prep.py for any additional data processing needs.

Evaluation

The model is evaluated using two primary metrics:

• Intersection over Union (IoU): Measures the overlap between predicted and actual masks.
• Dice Coefficient: Provides a score for similarity between predicted and actual masks, especially useful in segmentation tasks.

Conclusion

This project demonstrates how U-Net can be used for image segmentation tasks, providing scripts for preprocessing, training, and evaluating the model.