Keras UNet Architectures

This repository contains various U-Net architecture variants implemented in Keras for image segmentation tasks. These models provide different features and capabilities for building U-Net models quickly and efficiently.

Models Included

1. UNet2D: Basic 2D U-Net model.
2. ResUNet2D: 2D Residual U-Net model. Inspired from here.
3. ResUNetPlusPlus2D: 2D Residual U-Net++ model. Inspired from here.
4. AttentionUNet2D: 2D Attention U-Net model. Inspired from here.

Usage

Each model is implemented as a function that can be easily called to create the corresponding architecture. The models offer flexibility in terms of levels, convolutional layers per level, starting features, dropout rate, output and activation.

How to Use

1. Clone the repository.
2. Import the desired U-Net architecture function into your project.
3. Call the function with the appropriate parameters to create the model.

Example

from unet_models import create_unet_2D

# Create a 2D U-Net model
model = create_unet_2D(
    input_shape=(256, 256, 3),
    levels=4,
    convs_per_level=2,
    start_features=64,
    dropout=0.5,
    output_activation='sigmoid'
)

Contributions

Contributions to add more U-Net variants or improve existing ones are welcome! Please follow the standard contribution guidelines.

License

This project is licensed under the GNU General Public License v3.