DL Assignment 1 – CNN and Transfer Learning

This repository contains the implementation and report for a deep learning assignment focused on image classification using Convolutional Neural Networks (CNNs) and transfer learning.

Task Description

The goal is to evaluate the effectiveness of transfer learning by training a CNN model on the Stanford Dogs dataset and evaluating it on the Cats vs Dogs dataset through multiple experiments.

Based on the Keras CNN tutorial, we performed four experiments:

1. Train from scratch on Cats vs Dogs (baseline).
2. Transfer learning: Freeze all layers except the output.
3. Transfer learning: Re-train the first two convolutional layers along with the output.
4. Transfer learning: Re-train the last two convolutional layers along with the output.

Repository Structure

• notebook.ipynb – Implementation of all four experiments in Keras.
• report.pdf – Summary of results and analysis.
• README.md – This file.
• .gitignore - Ignores datasets, model checkpoints, etc.

Note: Datasets are large and excluded from version control via .gitignore.

Required Libraries

Please ensure the following libraries are installed in your Python environment:

• tensorflow
• keras
• matplotlib
• numpy
• pandas
• scikit-learn

How to Run

1. Install the dependencies:

pip install tensorflow keras matplotlib numpy pandas scikit-learn

2. Open notebook.ipynb using Jupyter Notebook, JupyterLab, or VS Code, and run all cells sequentially to reproduce the experiments.

Author

Arian Ghadirzadeh
Deep Learning Course (TDIS22) – Assignment 1
Jönköping University, School of Engineering (JTH)
April 2025