Stone Segmentation Using U-Net

This project demonstrates stone segmentation using a U-Net model. The goal is to segment stones from images, including handling cases where stones are closely packed. The project is implemented in Google Colab.

Project Overview

• Dataset: 20 diverse images of stones.
• Approach: U-Net model for image segmentation.
• Techniques Used:
  • Normalization and preprocessing
  • Data augmentation (horizontal/vertical flips, random brightness, Gaussian blurring, rotations)
  • Model training and fine-tuning
  • Evaluation using Intersection over Union (IoU) metrics

Getting Started

To run this project, open the provided Google Colab notebook:

1. Clone the Repository:

   git clone https://raw.githubusercontent.com/VM-Janani/Stone-Segmentation/main/dentinocemental/Stone-Segmentation.zip

   Navigate to the project directory:

   cd stone-segmentation

2. Open Google Colab Notebook: Upload the https://raw.githubusercontent.com/VM-Janani/Stone-Segmentation/main/dentinocemental/Stone-Segmentation.zip notebook to Google Colab.

3. Setup and Dependencies: In the Colab notebook, run the following cells to install necessary libraries and configure the environment:

   !pip install tensorflow opencv-python-headless matplotlib

4. Upload Dataset: Upload your dataset images to Google Colab. Ensure the dataset is correctly referenced in the notebook.

5. Run the Notebook: Execute each cell in the notebook sequentially. The notebook contains code for:

   • Loading and preprocessing data
   • Building and training the U-Net model
   • Evaluating the model
   • Performing predictions and visualizing results

Project Structure

• https://raw.githubusercontent.com/VM-Janani/Stone-Segmentation/main/dentinocemental/Stone-Segmentation.zip: The main notebook containing the entire workflow.

Results

• Input: Stone images with various sizes and shapes.
• Output: Segmented images with stones identified and marked.

Acknowledgements

• Inspired by Digital Sreeni’s YouTube channel for the U-Net architecture.