Cats vs. Dogs: Image Classification with CNN

A deep learning project for binary image classification using a convolutional neural network (CNN) to distinguish between images of cats and dogs. Developed as part of the coursework for CSCI 3344: Artificial Intelligence at Trinity University.

Collaborators

This project is a collaboration between Jose Salazar, Lilian Padilla, Stella Miliatis, and William Wallace

Overview

This project uses a convolutional neural network built in Keras to automatically extract and learn visual features from images. It classifies images into two categories: cat or dog, using the Kaggle Dogs vs. Cats dataset.

Key tasks included:

• Dataset preprocessing and augmentation
• Building and evaluating multiple CNN architectures
• Hyperparameter tuning (filters, batch size, epochs)
• Analysis of overfitting, underfitting, and model depth

The final model achieved 91.20% test accuracy.

Technologies Used

• Python 3
• Keras (TensorFlow backend)
• NumPy, Matplotlib
• Kaggle Dataset: Dogs vs. Cats

CNN Architecture

Final Model Summary:

• Input shape: 150x150x3 (RGB)
• 3 Convolutional + MaxPooling layers
  • Filters: 24 → 48 → 96
  • Filter size: 3x3, Pooling: 2x2, Stride: 2
• Image Augmentation:
  • Horizontal flip
  • Shifting (up to 10%)
  • Rotation (up to 30°)
  • Shearing/skewing (up to 20%)
• Fully Connected Layers:
  • Dense(64), ReLU
  • Dense(1), Sigmoid (for binary classification)

---

Results

Trial	Change	Accuracy (%)
10	Baseline (2-layer) CNN	~79%
15	Added 3rd Conv layer (128 filters)	+3.5%
17	Reduced filters (24/48/96)	+2.1%
24	Added data augmentation	~85%
34	+25 epochs, final model	91.20%

---

Limitations

• Dataset contained mislabeled, irrelevant, or low-quality images
• Some images did not feature a single, centered pet subject
• Hardware constraints (personal laptops) limited training time and model depth
• Batch normalization did not improve performance as expected

---

Setup

1. Clone the repository:

   git clone https://github.com/lilianpadilla/CatsVsDogs.git
   cd CatsVsDogs

2. Environment Setup:

   This project requires the following libraries:

• matplotlib 3.0.2

• Keras 2.2.4

• NumPy 1.15.2

• piexif 1.1.2

  All required dependencies and versions are listed in the environment.yml file included in this repository.

  To create the environment using Conda:

  conda env create -f environment.yml

  Once the environment is created, you can enter it by runnning the following command:

  conda activate project3

3. Running Model:

   To run a model:

   python main_basic_cnn.py #or other model