Fine-Tuning DenseNet-201 using FastAI

This repository demonstrates how to fine-tune a DenseNet-201 deep learning model for image classification using the FastAI library. The core implementation is contained within the finalModel.ipynb Jupyter Notebook.

📖 Project Overview

The primary goal of this project is to showcase transfer learning by adapting a pre-trained DenseNet-201 model (originally trained on ImageNet) to a new dataset. FastAI is utilized to streamline the data loading, model creation, and training loop.

Key Features

• Model Architecture: DenseNet-201
• Framework: FastAI (PyTorch backend)
• Technique: Transfer Learning (Fine-Tuning)

🚀 Getting Started

Follow these instructions to get a copy of the project up and running on your local machine.

Prerequisites

• Python 3.6+
• Jupyter Notebook or JupyterLab

Installation

1. Clone the repository

   git clone https://github.com/arupa444/Fine-Tuning-Densenet201-using-FastAI.git
   cd Fine-Tuning-Densenet201-using-FastAI

2. Install dependencies It is recommended to use a virtual environment. Install standard data science libraries and FastAI:

   pip install fastai jupyterlab

   (Note: FastAI will automatically install the compatible version of PyTorch)

💻 Usage

1. Launch Jupyter Lab:

   jupyter lab

2. Open finalModel.ipynb.
3. Run the notebook cells sequentially to observe the data loading, model setup, and training process.

📊 Notebook Highlights (finalModel.ipynb)

The notebook covers the following critical steps:

• DataBlock API: Flexible data loading and augmentation pipelines.
• Vision Learner: Initializing densenet201 with pretrained weights.
• lr_find(): FastAI's utility to find the optimal learning rate.
• fine_tune(): The one-cycle training policy for efficient transfer learning.
• Interpretation: Using ClassificationInterpretation to view confusion matrices and top losses.

🤝 Contributing

Contributions, issues, and feature requests are welcome!

📝 License

This project is open-source.