Image Segmentation

Overview

This project implements an image segmentation model using the U-Net architecture on images from the dashboard of Self-driving cars. U-Net is a convolutional neural network (CNN) primarily used for biomedical image segmentation but can also be applied to various segmentation tasks.

Features

• Data Preprocessing: Reads and processes images and masks for training.
• U-Net Model Implementation: Custom implementation of the U-Net architecture using TensorFlow/Keras.
• Training and Evaluation: Model training and visualization of segmentation results.
• Prediction Visualization: Displays model predictions for qualitative analysis.

Requirements

Ensure you have the following dependencies installed:

pip install tensorflow numpy pandas matplotlib imageio

Usage

1. Clone the Repository

git clone <repository_url>
cd <repository>

2. Run Jupyter Notebook

Execute the notebook to train and evaluate the model:

jupyter notebook Image_segmentation_Unet_v2.ipynb

3. Utility Functions

Data Processing

The utils.py file includes functions for:

• Loading images and masks
• Preprocessing images (resizing, normalization, etc.)

Model Definition

U-Net model implementation is available in utils.py:

from utils import unet_model
model = unet_model()

Display Results

from utils import display

display([input_image, true_mask, predicted_mask])

Results

The notebook provides visualizations of segmented outputs, including comparisons between ground truth masks and predictions.

Future Work

• Improve model performance using data augmentation.
• Experiment with different loss functions.
• Extend to multi-class segmentation tasks.

License

This project is licensed under the MIT License.

---

Author: Chirag N Vijay