YOLO Dog Breed Detection Web App

Table of Contents

• Overview
• Features
• Demo
• Installation
  • Prerequisites
  • Clone the Repository
  • Set Up the Environment
  • Install Dependencies
• Usage
  • Running the Application
  • Interacting with the Web App
• Metrics Handling
• Project Structure
• Troubleshooting
• Contributing
• License
• Contact

Overview

The YOLO Dog Breed Detection Web App is a powerful and user-friendly tool designed to detect and classify dog breeds within uploaded images using the YOLO (You Only Look Once) object detection framework. Leveraging a pre-trained YOLO model with 120 dog breed classes, this web application provides real-time detection results, including bounding boxes, confidence scores, and comprehensive validation metrics.

Features

• Real-Time Detection: Upload an image and receive instant detection results with bounding boxes and confidence scores.
• Comprehensive Metrics: View both overall and per-class validation metrics to assess model performance.
• User-Friendly Interface: Built with Gradio for an intuitive and interactive user experience.
• Scalable Architecture: Designed to handle a large number of classes efficiently.
• High Accuracy: Utilizes a robust YOLO model trained on a diverse dog breed dataset.

Dataset

https://universe.roboflow.com/iliescu-mihail-doirn/stanford-dogs-dataset-dog-breed/dataset/1

Demo

Screenshot demonstrating the YOLO Dog Breed Detection Web App in action.

Installation

Prerequisites

• Python 3.10.12 or higher
• CUDA 12.1 (for GPU acceleration) (Optional but recommended for faster inference)
• Git (to clone the repository)

Clone the Repository

git clone https://github.com/yourusername/your-repo-name.git
cd your-repo-name

Set Up the Environment

It's recommended to use a virtual environment to manage dependencies.

Using venv:

python -m venv venv

Activate the Virtual Environment:

• Windows:

  venv\Scripts\activate

• macOS and Linux:

  source venv/bin/activate

Install Dependencies

Ensure you have pip updated to the latest version:

pip install --upgrade pip

Install the required Python packages using the provided requirements.txt:

pip install -r requirements.txt

Note: The torch package specified in requirements.txt is built with CUDA 12.1 support. If your system has a different CUDA version or if you're using a CPU-only setup, adjust the PyTorch installation accordingly. Visit the PyTorch Official Installation Page for the appropriate command.

For example, for CPU-only support:

pip install torch==2.5.1 torchvision==0.15.2 torchaudio==2.5.1

Usage

Running the Application

Ensure that the best.pt YOLO model file and the data.yaml configuration file are present in the project directory.

Execute the Python script to launch the web application:

python app.py

Replace app.py with the actual filename if different.

Interacting with the Web App

1. Access the Web Interface:

   After running the script, Gradio will provide a local URL (e.g., http://127.0.0.1:7860/). Open this URL in your web browser.

2. Upload an Image:

   • Click on the "Upload Image" button.
   • Select an image of a dog from your local machine.

3. Run Inference:

   • Click the "Run Inference" button.
   • The application will process the image, detect dog breeds, and display the results.

4. View Results:

   • Annotated Image: Displays the uploaded image with bounding boxes and labels indicating detected breeds and confidence scores.
   • Detection Results: A table listing detected classes with their confidence scores and bounding box coordinates.
   • Validation Metrics: A table showing overall and per-class metrics to evaluate model performance.

Metrics Handling

The application displays both overall and per-class metrics to provide a comprehensive view of the model's performance.

• Overall Metrics:

  • Precision: The ratio of true positive detections to the total predicted positives.
  • Recall: The ratio of true positive detections to the actual positives.
  • mAP50: Mean Average Precision at 50% IoU threshold.
  • mAP50-95: Mean Average Precision across IoU thresholds from 50% to 95%.

• Per-Class Metrics:

  • Detailed metrics for each of the 120 dog breed classes, including Precision, Recall, mAP50, and mAP50-95.

Note: Metrics are precomputed and hardcoded into the application for demonstration purposes. For real-world applications, consider implementing dynamic metric computation.

Project Structure

your-repo-name/
│
├── app.py                   # Main application script
├── best.pt                  # Pre-trained YOLO model
├── data.yaml                # YOLO data configuration file
├── requirements.txt         # Python dependencies
├── README.md                # Project documentation

Troubleshooting

• Model File Not Found:

  Ensure that best.pt is located in the project directory or update the model_path variable in app.py to the correct location.

• CUDA Errors:

  If you encounter CUDA-related errors, verify that your system has the appropriate CUDA version installed. If not using a GPU, adjust the PyTorch installation to CPU-only as mentioned in the Installation section.

• Dependencies Issues:

  Ensure all dependencies are installed correctly. You can reinstall them using:

  pip install --force-reinstall -r requirements.txt

• Gradio Not Launching:

  Check if the required ports are free and not blocked by firewalls. You can specify a different port by modifying the demo.launch() line:

  demo.launch(server_port=7861)

License

This project is licensed under the MIT License.

Contact

For any questions or suggestions, please contact:

• Name: Partha Pratim Ray
• Email: [email protected]

---

Hugging Face Space

https://huggingface.co/spaces/csepartha/yolo11_stanford_dog_detection Happy Detecting! 🐶🔍