Image Segmentation Pipeline Using DeepLabV3

This repository contains an Image Segmentation Application built with DeepLabV3, leveraging PyTorch and Streamlit for a seamless and interactive experience. The application processes images to perform segmentation tasks, providing fine-tuned results with performance metrics such as Intersection over Union (IoU).

Features

• DeepLabV3 Model: Utilizes a pre-trained DeepLabV3-ResNet101 model for semantic segmentation.
• Interactive Interface: Built with Streamlit, enabling parameter tuning and visualization.
• Customizable Segmentation: Adjust kernel size and iterations for morphological operations.
• IoU Score Calculation: Evaluate segmentation accuracy against ground truth masks.
• Dataset Support: Supports datasets with original images and corresponding no-background masks.

Requirements

Ensure you have Python 3.8 or higher installed. The required Python libraries can be installed using the provided requirements.txt.

pip install -r requirements.txt

Usage

1. Clone the repository:

   git clone https://github.com/yourusername/image-segmentation-pipeline.git
   cd image-segmentation-pipeline

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Streamlit app:

   streamlit run image_segmentation_app.py

4. Access the application in your browser at http://localhost:8501.

Input Parameters

• Number of Images: Select the number of images to process and display.
• Kernel Size: Define the size of the kernel for morphological operations.
• Iterations: Specify the number of iterations for refining segmentation masks.

Directory Structure

image-segmentation-pipeline/
├── public_hand_dataset/           # Dataset directory (default)
│   ├── [image_id]/
│       ├── original/              # Original images
│       └── no_bg/                 # Corresponding no-background masks
├── image_segmentation_app.py      # Main application file
├── requirements.txt               # Dependencies
└── README.md                      # Project documentation

Dataset

The app expects the dataset to be structured as follows:

public_hand_dataset/
├── [image_id]/
│   ├── original/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   └── no_bg/
│       ├── image1.jpg
│       ├── image2.jpg

Place your dataset in the public_hand_dataset/ directory or specify a custom path in the code.

Key Functions

• load_model(): Loads the pre-trained DeepLabV3 model.
• load_dataset(dataset_dir): Loads image paths from the dataset.
• process_images(file_paths, kernel_size, iterations, model): Processes images and applies segmentation.
• fn_compute_iou(predicted, ground_truth): Computes the IoU score for evaluation.

Results Visualization

The app displays the following:

1. Original Image: Input image for segmentation.
2. Initial Segmentation: Initial segmentation mask with IoU score.
3. Fine-Tuned Segmentation: Refined segmentation mask with IoU score.
4. Expected Segmentation: Ground truth segmentation mask.

Technologies Used

• Python: Primary language for implementation.
• PyTorch: For deep learning model loading and inference.
• Streamlit: For creating an interactive web app.
• OpenCV: For morphological operations and image processing.
• Pillow: For image manipulation and preprocessing.

Contributions

Contributions are welcome! If you'd like to contribute, please fork the repository and create a pull request.

Acknowledgments

• DeepLabV3: Pre-trained model from PyTorch's TorchVision library.
• Streamlit: Simplifies building data-driven apps.