EDTH-Warsaw-shahed136-detector

YOLO12 Shahed-136 Drone Detector

<p align="center"> <img src="https://img.shields.io/badge/YOLO-12-blue" alt="YOLO12"> <img src="https://img.shields.io/badge/PyTorch-2.0+-red" alt="PyTorch"> <img src="https://img.shields.io/badge/License-MIT-green" alt="License"> </p>

Real-time drone detection model built for the European Defense Hackathon Warsaw 2024 by Imperial College London SPQR Team.

Model Description

This model detects Shahed-136 military drones in real-time using YOLO12's attention-centric architecture. YOLO12 introduces Area Attention mechanisms and R-ELAN feature aggregation for improved accuracy while maintaining real-time inference speeds.

Quick Start

Installation

pip install ultralytics

Inference

from ultralytics import YOLO

# Load model
model = YOLO("path/to/best.pt")

# Run inference
results = model("image.jpg")

# Show results
results[0].show()

Webcam Detection

from ultralytics import YOLO
import cv2

model = YOLO("path/to/best.pt")
cap = cv2.VideoCapture(0)

while True:
    ret, frame = cap.read()
    if not ret:
        break
    
    results = model(frame, conf=0.25, verbose=False)
    annotated = results[0].plot()
    
    cv2.imshow("Drone Detection", annotated)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

Download Specific Version

from huggingface_hub import hf_hub_download

# Download from specific branch (training run)
model_path = hf_hub_download(
    repo_id="shng2025/EDTH-Warsaw-shahed136-detector",
    filename="best.pt",
    revision="yolo12m_20251206_144537",  # Branch name = run name
)

Training

Dataset

• Source: Synthetic renders + real footage of Shahed-136 drones
• Images: ~8,000 labeled images
• Format: YOLOv8/YOLO12 (Ultralytics)
• Split: 80% train / 15% val / 5% test

Hyperparameters

Training Command

from ultralytics import YOLO

model = YOLO("yolo12m.pt")
model.train(
    data="data.yaml",
    epochs=100,
    batch=16,
    imgsz=640,
    optimizer="AdamW",
    lr0=0.001,
    augment=True,
    mosaic=1.0,
)

Branches

Each training run is saved to a separate branch:

Experiment Tracking

Training runs are logged to Weights & Biases:

• Entity: Imperial-College-London-SPQR
• Project: European-Defense-Hackathon-Warsaw
• View Dashboard

Roadmap

• Single-class drone detection (Shahed-136)
• Multi-class detection (drones, humans, vehicles, birds)
• Indoor environment adaptation (office, chairs, hands)
• Edge deployment optimization (TensorRT, ONNX)
• Integration with tracking systems

Limitations

• Trained primarily on synthetic/rendered data
• Single class detection only (current version)
• May have reduced accuracy in:
  • Heavy fog/rain conditions
  • Night/low-light scenarios
  • Extreme viewing angles
  • Cluttered backgrounds (trees, birds)

Citation

@software{edth_shahed136_detector,
  author = {EDTH Team - Orzel 1},
  title = {YOLO12 Shahed-136 Drone Detector},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/shng2025/EDTH-Warsaw-shahed136-detector}
}

@article{yolo12,
  title = {YOLO12: Attention-Centric Real-Time Object Detectors},
  author = {Tian, Yunjie and Ye, Qixiang and Doermann, David},
  journal = {arXiv preprint arXiv:2502.12524},
  year = {2025}
}

License

MIT License - See LICENSE for details.

Team

Imperial College London SPQR - European Defense Hackathon Warsaw 2024

---

<p align="center"> <i>Built for defense. Built for safety.</i> </p>