YOLOv8m_defence

YOLOv8m Defence

Model Overview

YOLOv8m Defence is a specialized object detection model fine-tuned from the ultralytics/YOLOv8 checkpoint for defence and transportation applications. The model was trained for over 50 epochs on a private defence-related dataset to detect 18 categories of aircraft, vehicles, and ships with high accuracy and low latency requirements.

A deployment of the model is available at Hugging Face Spaces for demonstration of its capabilities.

Model Details

• Model Type: Object Detection
• Base Architecture: YOLOv8m
• Framework: PyTorch
• Training Epochs: 50+
• Number of Classes: 18
• Input: RGB Images
• Output: Bounding boxes with class predictions and confidence scores

Supported Classes

The model detects the following 18 object categories:

Performance

The model achieved the following performance metrics in its original evaluation environment:

• [email protected]:0.05:0.95: 0.598
• Speed Score: 0.933

Note: These scores were achieved with the fully optimized TensorRT version. The PyTorch model provided here may have different performance characteristics.

Usage

Installation

pip install ultralytics

Quick Start

from ultralytics import YOLO
from PIL import Image

# Load the model
model = YOLO('path/to/yolov8m_defence.pt')

# Run inference
results = model('path/to/your/image.jpg')

# Process results
for r in results:
    print(f"Detected {len(r.boxes)} objects")
    
    for box in r.boxes:
        # Get bounding box coordinates
        x1, y1, x2, y2 = box.xyxy[0]
        bbox = [int(x1), int(y1), int(x2 - x1), int(y2 - y1)]
        
        # Get class and confidence
        class_id = int(box.cls[0])
        confidence = float(box.conf[0])
        class_name = model.names[class_id]
        
        print(f"  {class_name} (ID: {class_id}): {confidence:.2f} at {bbox}")

Batch Inference

# Run inference on multiple images
results = model(['image1.jpg', 'image2.jpg', 'image3.jpg'])

for i, r in enumerate(results):
    print(f"Image {i+1}: {len(r.boxes)} detections")

Visualization

# Display results with bounding boxes
results = model('image.jpg')
annotated_image = results[0].plot()

# Convert BGR to RGB for display
from PIL import Image
import numpy as np
Image.fromarray(annotated_image[..., ::-1]).show()

Training Details

Dataset

• Source: Private defence dataset (proprietary)
• Classes: 18 object categories
• Augmentations: Mosaic, flips, color adjustments
• Annotations: Bounding box format

Training Configuration

• Base Model: ultralytics/yolov8m
• Training Epochs: 50+
• Framework: Ultralytics YOLO
• Optimization: Standard YOLOv8 training pipeline

Post-Training Optimization

The original model underwent additional optimization including:

• Model pruning
• Quantization
• TensorRT conversion (for deployment)

This repository contains the pre-optimization PyTorch model for maximum compatibility and ease of use.

Intended Use Cases

Primary Applications

• Military and defence object detection
• Transportation vehicle monitoring
• Surveillance and reconnaissance
• Aerial and maritime asset identification

Suitable For

• Direct inference on defence-related imagery
• Transfer learning for similar detection tasks
• Baseline model for military/civilian vehicle detection
• Research and development in computer vision

Limitations

• Scope: Only detects the 18 trained object categories
• Domain: Performance may vary on images significantly different from training data
• Speed: This PyTorch version is slower than the optimized TensorRT variant
• Hardware: No specific GPU requirements, but GPU acceleration recommended

Ethical Considerations

This model is designed for defence and security applications. Users should:

• Ensure compliance with local laws and regulations
• Consider privacy implications when processing imagery
• Use responsibly and ethically in surveillance applications
• Respect international laws regarding military technology

Citation

@misc{yolov8_ultralytics,
  author = {Jocher, Glenn and Chaurasia, Ayush and Qiu, Jing},
  title = {YOLO by Ultralytics},
  year = {2023},
  howpublished = {\url{https://github.com/ultralytics/ultralytics}},
}

License

This model is released under the Apache 2.0 License.

Model Card Contact

For questions about this model card or the model itself, please refer to the repository issues or contact the model authors.