Dog Breed Image Classifier Project

This project is designed to classify dog breeds from images using a pre-trained deep learning model. It includes scripts for training, evaluating, and predicting using models like AlexNet, ResNet, and VGG. Additionally, the project compares performance across different models on various datasets.

Project Overview

The project aims to:

• Classify images of pets (dogs) using deep learning.
• Use transfer learning by leveraging pre-trained models from torchvision.models.
• Provide a command-line interface for flexible model usage.

Installation

Dependencies

To run this project, the following Python libraries are required:

• torch
• argparse
• PIL
• os
• json
• shutil

Install dependencies with:

pip install torch pillow

Project Files

Core Scripts

• check_images.py: Verifies image file structure and formats.
• classify_images.py: Classifies images using the chosen deep learning model.
• get_input_args.py: Parses command-line arguments, including model architecture and file paths.
• get_pet_labels.py: Extracts labels (breeds) from image filenames.
• adjust_results4_isadog.py: Adjusts classification results to account for whether the image is a dog.
• calculates_results_stats.py: Computes statistics (accuracy, precision, recall, etc.) based on model predictions.
• print_results.py: Prints out a summary of classification results and statistics.
• print_functions_for_lab_checks.py: Helper functions for debugging and validation.

Additional Files

• run_models_batch.sh: Batch script to run model training and classification for pet images.
• run_models_batch_uploaded.sh: Batch script for running models on uploaded images.
• imagenet1000_clsid_to_human.txt: Maps ImageNet class IDs to human-readable labels.
• dognames.txt: Contains a list of dog breeds recognized by the classifier.

Pretrained Model Results

• alexnet_pet-images.txt: Classification results using the AlexNet model on pet images.
• resnet_pet-images.txt: Classification results using the ResNet model on pet images.
• vgg_pet-images.txt: Classification results using the VGG model on pet images.
• alexnet_uploaded-images.txt: Classification results using the AlexNet model on uploaded images.
• resnet_uploaded-images.txt: Classification results using the ResNet model on uploaded images.
• vgg_uploaded-images.txt: Classification results using the VGG model on uploaded images.

Folders

• pet_images/: Contains the pet images for model training and testing.
• uploaded_images/: Contains the uploaded images used for classification testing.

Usage

Command Line Arguments

You can run the scripts from the command line, with the following customizable arguments:

1. Classify Images

   python classify_images.py --dir <path_to_images> --arch <model_architecture> --dogfile dognames.txt

   Example:

   python classify_images.py --dir pet_images/ --arch vgg --dogfile dognames.txt

2. Check Images

   python check_images.py --dir <path_to_images>

3. Run Models Batch

   ./run_models_batch.sh

Options for Command Line:

• --dir: Directory path to the images.
• --arch: Model architecture to use (alexnet, resnet, vgg).
• --dogfile: Text file containing a list of recognized dog breeds.

Results

This project tests the accuracy of different pre-trained models on pet images, specifically dogs. It evaluates the model's ability to correctly classify whether the image is a dog, and if so, identifies the breed.

Key metrics such as precision, recall, and F1 score are calculated and printed using the print_results.py script.