YOLO Hotdog Classifier

Claude wrote this read.me and did some of the work because I'm lazy. Feel free to let me know if you have issues

A computer vision model that classifies images as either "hotdog" or "not hotdog", inspired by the Silicon Valley TV show. This project uses YOLOv8 for image classification.

Project Description

This classifier uses a fine-tuned YOLOv8 model to determine whether an image contains a hotdog or not. The model achieves approximately 90.6% accuracy on the validation dataset and has a small footprint of only 3.0MB, making it suitable for Sony IMX500 onboard AI accelerator.

Features

• Binary image classification (hotdog/not hotdog)
• Fast inference (~1.5ms per image on CPU)
• Small model size (~3.0MB)
• Simple training and inference pipeline

Dataset Information

The dataset consists of:

• Training dataset: 498 images (balanced between hotdogs and non-hotdogs)
• Validation dataset: 100 images

Dataset Structure

The dataset is organized as follows:

dataset/
├── train/
│   ├── hotdog/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   │   └── ...
│   └── not_hotdog/
│       ├── image1.jpg
│       ├── image2.jpg
│       └── ...
└── val/
    ├── hotdog/
    │   ├── image1.jpg
    │   ├── image2.jpg
    │   └── ...
    └── not_hotdog/
        ├── image1.jpg
        ├── image2.jpg
        └── ...

Each folder contains images of the respective class for training and validation.

Dataset Preparation

The repository includes a split_dataset.py script that helps you prepare your dataset by:

1. Splitting your dataset into training and validation sets
2. Organizing images into the required folder structure
3. Balancing classes if needed

Usage:

python split_dataset.py --source your_images_folder --train_ratio 0.8

Data Configuration

The dataset configuration is specified in data.yaml:

# Dataset paths
path: ./dataset  # Root directory
train: train/    # Training data relative to path
val: val/        # Validation data relative to path

# Classes
names:
  0: hotdog
  1: not_hotdog

Installation and Setup

Requirements

• Python 3.8 or higher
• PyTorch 1.10 or higher
• Ultralytics YOLO package

Installation

1. Clone this repository:

   git clone https://github.com/Ruxpins/yolo-hotdog-classifier.git
   cd yolo-hotdog-classifier
   

2. Install the required packages using requirements.txt:

   pip install -r requirements.txt
   

Project Structure

yolo_hotdog_classifier/
├── data.yaml         # Dataset configuration
├── train.py          # Training script
├── predict.py        # Inference script
├── split_dataset.py  # Dataset preparation tool
├── requirements.txt  # Dependencies
├── dataset/          # Dataset directory
│   ├── train/        # Training images
│   └── val/          # Validation images
└── runs/             # Training results and weights

Usage

Training

To train the model with default parameters:

python train.py

The script automatically:

• Loads the dataset from the dataset directory
• Initializes a YOLOv8n classification model
• Trains for 50 epochs with batch size 16
• Saves the best model weights to runs/classify/hotdog_classifier/weights/best.pt

You can customize training parameters by modifying train.py.

Inference

To classify new images:

python predict.py --source path/to/your/image.jpg

Or for multiple images or a directory:

python predict.py --source path/to/directory

Model Performance

• Accuracy: 90.6% on validation set
• Inference speed: ~1.5ms per image on CPU
• Model size: ~3.0MB

Hardware Support

The model can run on:

• CPU (default)
• NVIDIA GPU (with CUDA)
• AMD GPU (with ROCm support - requires additional setup)

License

[Insert your license information here]

Acknowledgements

• Ultralytics for the YOLO implementation
• Claude wrote this read.me because I was too lazy. Please let me know if there are any errors. I was not able to find any beyond what I've already fixed.
• Original dataset came from https://www.kaggle.com/datasets/dansbecker/hot-dog-not-hot-dog