https://github.com/hoverslam/image-segmentation

Multi-class image segmentation using a U-Net architecture.
https://github.com/hoverslam/image-segmentation

image-segmentation pytorch unet

Last synced: 12 days ago
JSON representation

Multi-class image segmentation using a U-Net architecture.

• Host: GitHub
• URL: https://github.com/hoverslam/image-segmentation
• Owner: hoverslam
• License: mit
• Created: 2024-08-28T12:23:12.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2024-09-11T06:08:28.000Z (2 months ago)
• Last Synced: 2024-10-11T07:43:18.736Z (about 1 month ago)
• Topics: image-segmentation, pytorch, unet
• Language: Jupyter Notebook
• Homepage:
• Size: 11 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Image Segmentation

This repository contains a deep learning project focused on performing image segmentation on the Oxford-IIIT Pet Dataset. Utilizing a U-Net architecture, the model is trained to accurately segment pet images, identifying the precise boundaries of the animals within the images. This project serves as a practical example of implementing U-Net for semantic segmentation tasks in computer vision.

## Installation

1. Clone the repository:

```

git clone https://github.com/hoverslam/image-segmentation

```

2. Navigate to the directory:

```

cd image-segmentation

```

3. Set up a virtual environment:

```bash

# Create a virtual environment

python -3.11 -m venv .venv

# Activate the virtual environment

.venv\Scripts\activate

```

4. (Optional) Install PyTorch with CUDA support:

```

pip install torch==2.3.1 torchvision==0.18.1 --index-url https://download.pytorch.org/whl/cu118

```

5. Install the dependencies:

```

pip install -r requirements.txt

```

## U-Net

The U-Net architecture is a convolutional neural network designed for image segmentation. It consists of an encoder that captures context and a decoder that enables precise localization, making it highly effective for segmenting images. Skip connections are a key feature of U-Net that link the encoder and decoder pathways directly. They preserve high-resolution features and detailed spatial information by combining low-level details with high-level context.

> [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://arxiv.org/abs/1505.04597) by Olaf Ronneberger, Philipp Fischer, and Thomas Brox (2015).

## Dataset

The [Oxford-IIIT Pet Dataset](https://www.robots.ox.ac.uk/~vgg/data/pets/) is a collection of images featuring 37 different pet breeds. Each image comes with a corresponding segmentation mask that divides the image into three distinct classes: pet, background, and border. The dataset provides a total of 7349 images, split into 3680 training samples and 3669 test samples. Notably, the training data is further divided into training and validation sets using a 75:25 ratio.

## Results

| | Train | Val | Test |

|---|---|---|---|

| IoU | 0.7718 | 0.7013 | 0.6976 |

| Accuracy | 0.9507 | 0.9224 | 0.9181 |







## License

The code in this project is licensed under the [MIT License](LICENSE.txt).