Dogs and Wolves Classifier

This project implements a binary classifier to distinguish between dogs and wolves using a custom convolutional neural network (CNN) built with PyTorch.

Setup

Prerequisites

• Python 3.7 or higher
• PyTorch
• torchvision
• matplotlib
• numpy
• PIL

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/DogsAndWolves.git cd DogsAndWolves

2. Create and activate a virtual environment:

   python3 -m venv venv source venv/bin/activate

3. Install the dependencies:

   pip3 install torch torchvision matplotlib numpy pillow

Model training and evaluation

To train the model, run the following command:

python3 main.py train --data_path path/to/dataset 
(Default path is data/)

Make sure to change directory that specifies where the weights will be saved after training in train.py file (Line 130)

Predicting Images

To predict the class of images in a directory, run:

python3 main.py predict --image_path path/to/your/images/ --weights_path path/to/your/weights.pth
(Default for --image path is /newImages so you can add new images to this folder)
(Default for --weights_path is weights/old/final_model_weights.pth)

Project Components

models/
    cnn.py: Contains the definition of the CNN model.
    early_stopping.py: Implements early stopping to avoid overfitting.

scripts/

    train.py: Script for training the model.
    evaluate.py: Script for evaluating the model.
    predict.py: Script for predicting the class of images.

Notes

The project supports GPU acceleration. If you have a CUDA-compatible GPU, the training and evaluation will automatically use it. Otherwise, it will fall back to CPU

Authors

dis012