🎯 Object Detection with PyTorch & Flask

A complete end-to-end object detection system that leverages PyTorch's Faster R-CNN pretrained model and serves detection via a Flask web application and REST API. This project supports detecting objects in images and videos using a modular and extensible pipeline.

📸 Demo

🚀 Features

• 🔍 Object detection using Faster R-CNN ResNet50 FPN V2 with COCO weights
• 🖼️ Detect objects in images with bounding box visualization
• 🎥 Detect objects frame-by-frame in videos with FPS display
• 🌐 Web interface for easy image input and result visualization
• ⚙️ REST API backend for programmatic detection requests
• 💻 Modular Python scripts for model loading, inference, and visualization
• 🗂️ Organized project structure for ease of use and extension

📁 Project Structure

Object-Detection-With-PyTorch-and-Custom-the-Model-By-Flask
├── assets/                  # Static assets for UI or documentation
├── data/                    # Dataset files (annotations, images)
├── docs/                    # Documentation resources
├── input/                   # Input images/videos for testing
├── outputs/                 # Output detections and results
├── python/                  # Core Python modules
│   ├── detect_utils.py      # predict() + draw_boxes()
│   ├── model.py             # Load pretrained Faster R-CNN model
│   └── utils.py             # COCO class names
├── static/                  # Static files served by Flask
│   ├── css/                 # CSS styles
│   │   └── main.css         # Stylesheet for the web UI
│   └── uploads/             # Uploaded images directory
├── templates/               # HTML templates for Flask web pages
│   ├── base.html            # HTML base layout
│   └── homepage.html        # Home page with upload and detection UI
├── translations/            # Language translations (if applicable)
├── api_app.py               # Flask app and routes
├── detect_api.py            # Core detection for Flask app
├── detect_image.py          # CLI: detect objects in image
├── detect_video.py          # CLI: detect objects in video
├── LICENSE                  # License file (MIT)
├── README.md                # This README file
└── requirements.txt         # Python dependencies

💡 Installation

1. Clone the repository:

git clone https://github.com/TareqAlKushari/Object-Detection-With-PyTorch-and-Custom-the-Model-By-Flask.git
cd Object-Detection-With-PyTorch-and-Custom-the-Model-By-Flask

2. (Optional) Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install required packages:

pip install -r requirements.txt

🎯 Usage

Image Detection

Detect objects on a single image and save the output with bounding boxes.

python detect_image.py path/to/image.jpg --threshold 0.5

The output image will be saved in outputs/ directory.

Video Detection

Detect objects on a video file (or webcam stream) frame-by-frame with FPS display.

python detect_video.py path/to/video.mp4 --threshold 0.5

Processed video will be saved in outputs/ directory.

Press q to quit the video window early.

Flask Web Application

Start the interactive web app to upload images and visualize detection results in the browser.

python api_app.py

Open http://localhost:9000 in your browser.

• Input the path to an image accessible to the server for detection.
• View the annotated image rendered in the web page.

REST API Integration

Use detect_api.py functions to integrate detection into other applications or create REST endpoints.

Example curl request (assuming you add an API endpoint):

curl -X POST -F image=@path/to/image.jpg http://localhost:5000/predict

🧠 How It Works

🔍 Backend Inference Flow

• python/model.py: Loads pretrained model from torchvision.models.detection

• detect_utils.py:

  • predict(image, model, device, threshold) – preprocesses image, runs inference, filters predictions
  • draw_boxes() – draws bounding boxes and class labels using OpenCV

• utils.py: Defines the 91 COCO classes used for label mapping

🎨 Web UI Logic

• HTML/CSS in templates/ and static/css/

• Input form allows user to submit image path

• api_app.py:

  • Renders homepage.html
  • Calls detect_api.py to run detection on the input
  • Saves and displays output via HTML <img>

🧾 COCO Classes (Examples)

• person, bicycle, car, motorcycle, airplane, bus
• dog, cat, horse, sheep, cow
• bottle, chair, laptop, keyboard, clock
• ... and more (total: 91 categories)

📷 Sample Output

Input	Detected Output

⚙️ Dependencies

• Python 3.7+
• PyTorch
• torchvision
• OpenCV
• Flask
• Pillow
• numpy

(See requirements.txt for full list.)

📄 License

This project is licensed under the MIT License.

👤 Author

Tareq Al Kushari 🔗 GitHub Profile

---

If you encounter any issues or have suggestions, please open an issue or submit a pull request. Happy detecting! 🚀