🪖 ArgusAI

Deep learning–powered object detection framework using YOLOv8 for defense and surveillance applications.

ArgusAI is a web-based intelligent vision system built with Flask and YOLOv8.
It allows users to upload images and automatically detect and classify various military and civilian vehicles and aircraft.
The model is trained on a custom defense dataset featuring six distinct object classes.
(Video inference functionality coming soon!)

---

🚀 Features

• 📸 Upload an image through an intuitive web interface
• 🤖 Object detection powered by YOLOv8 (Ultralytics)
• 🧩 Detects both civilian and military vehicles/aircraft
• 💾 Automatically processes and displays detection results
• 🧠 Modular design separating model logic and Flask routes
• ⚡ Real-time inference with bounding boxes and confidence scores

---

🧠 Model Information

• Model Architecture: YOLOv8 (PyTorch, Ultralytics)
• Trained Weights: best.pt
• Classes:
  • Civilian Aircraft
  • Civilian Car
  • Military Aircraft
  • Military Helicopter
  • Military Tank
  • Military Truck
• Framework: PyTorch
• Backend Integration: Flask

---

🔄 Application Flow

1. User Uploads Image

   • The image is sent from the HTML form to the Flask backend (app.py).
   • On form submission, a POST request is sent to the Flask backend (/predict route).
   • The uploaded image is accessed in Flask and saved to /uploads folder.

2. Flask Handles Request

   • After the image is saved the flask calls the YOLO inference function yolo_model.predict.
   • The annotated image path is returned and stored in the variable processed_image_path.

3. YOLO Inference

   • The predict(image_path) function in yolo_model.py is called and it performs detection.
   • Then after this a full output path is generated for locating the annotated image.
   • This full output image path is stored in processed_image_path as mentioned earlier.

4. Result Generation

   • The processed image with bounding boxes and class labels is saved to the runs/detect/predict/ folder.

5. Result Display

   • Flask dynamically loads and renders the processed image on the webpage using the result.html page.

---

🧰 Tech Stack

Component	Technology
Frontend	HTML, CSS (Jinja2 Templates)
Backend	Flask (Python)
Model	YOLOv8 (Ultralytics)
Framework	PyTorch
Inference	Custom YOLO model integrated with Flask

---

🧩 File Overview

app.py

• Initializes the Flask application
• Handles image upload and result rendering
• Connects the user interface with YOLO inference logic

yolo_model.py

• Contains all YOLOv8 inference code
• Loads model weights (best.pt) and performs detection
• Returns the path to the output image for display

requirements.txt

• Lists all Python dependencies (Flask, Ultralytics, Torch, OpenCV, etc.)

index.html

• Contains the code for landing page and image uploading.

result.html

• Contains the code for displaying the annotated image.

---

📦 Requirements

• ultralytics
• pandas
• numpy
• fastapi
• jinja2
• python-multipart

Running the model

The repository include the frontend code (index.html & result.html), backend code(app.py), Yolo training and inferencing code(yolo_model.py) as well as trained model weights i.e best.pt. Since the trained model weights (best.pt) are already included, users can directly run inference without retraining the model.

1. Clone the repository.
2. Then replace the model_path with the appropriate directory path.
3. Create a seperate uploads folder to save the uploaded images. In case you haven't, the app.py will generate the folder if it doesn't exist.
4. By default, the annotated image is stored in /runs/detect/predict due to save=True in yolo_model.py.