Semantic Segmentation with MobileNetV2 and U-Net

This project demonstrates semantic image segmentation using a custom architecture that combines MobileNetV2 and U-Net. The goal is to accurately segment objects in images, showcasing a robust approach to image analysis.

Project Overview

• The u-net-on-carvana.ipynb notebook contains the entire project pipeline, from data preprocessing to model training and evaluation.

Prerequisites

• Python (>=3.6)
• PyTorch (>=1.0)
• torchvision
• tqdm
• matplotlib
• numpy
• PIL

Project Structure

• u-net-on-carvana.ipynb: Jupyter Notebook containing the project code.
• MobileNetV2_Unet_wts.pth: Saved model weights.
• MobileNetV2_Unet_model.pth: Saved entire model (architecture and weights).

Getting Started

1. Clone this repository: git clone https://github.com/Rumit95/Semantic-Segmentation-of-Image
2. Navigate to the project directory: cd semantic-segmentation
3. Open u-net-on-carvana.ipynb using Jupyter Notebook or Jupyter Lab.
4. Run each cell sequentially to execute the project steps.

Usage

• Follow the steps in the notebook to preprocess data, create the custom MobileNetV2 U-Net model, train the model, and visualize results.
• You can modify hyperparameters, the number of training epochs, or experiment with different architectures to suit your needs.

Model Architecture

• Model Architecture

Credits

• The MobileNetV2 architecture is based on the original paper by Sandler et al. [https://arxiv.org/abs/1801.04381].
• U-Net architecture reference: Olaf Ronneberger, Philipp Fischer, Thomas Brox. "U-Net: Convolutional Networks for Biomedical Image Segmentation."

License

This project is licensed under the MIT License - see the LICENSE file for details.

Feel free to reach out if you have any questions or suggestions!