TinyVGG-inspired CNN Binary Classifier

This project implements a Convolutional Neural Network (CNN) binary classifier inspired by the TinyVGG architecture. The model is designed for binary classification tasks where the goal is to categorize input images into two classes.

Table of Contents

• Model Architecture
• Requirements
• Usage
• Training and Evaluation
• Contributor

Model Architecture

The CNN model is based on the TinyVGG architecture with modifications for binary classification. The architecture consists of:

• Multiple convolutional layers followed by max-pooling.
• ReLU activation functions after each convolution.
• Fully connected layers at the end to output a binary classification decision.

Key Features:

• Lightweight model, suitable for small datasets.
• Two output nodes using softmax activation for binary classification.

Architecture Overview:

Conv2D -> ReLU -> MaxPooling
Conv2D -> ReLU -> MaxPooling
Flatten -> Fully Connected -> ReLU
Fully Connected -> Output (Softmax for binary classification)

Requirements

To run this project, the following packages are required:

• Python 3.x
• PyTorch
• OpenCV (for image preprocessing)
• NumPy
• Matplotlib (for visualizations)
• tqdm (for progress tracking)

You can install the required packages using pip:

pip install torch torchvision opencv-python numpy matplotlib tqdm

Usage

1. Data Preparation:

   • Place your training and validation images in respective folders: datasets/train/ and datasets/val/, with subfolders for each class (e.g., class0/, class1/).

2. Dataset:

   • For this project, you can use the dataset from Kaggle: Binary Image Classification Dataset

Training and Evaluation

To the train, evaluate and test the model, make necessary changes and run the main script.

• The script will output accuracy, loss, and other metrics for evaluation.

Contributor

This project is developed by B M Manohara @Manohara-Ai

---