🐺 CerberusFace-YOLOv8: High-Accuracy Face Detection

CerberusFace-YOLOv8 is a fine-tuned face detection model based on YOLOv8 Nano, optimized for detecting small, distant, or low-resolution faces in challenging real-world conditions (CCTV, surveillance, crowd monitoring, etc.).

Trained on the WIDER FACE dataset, this model balances speed and accuracy, with added features like real-time zoom, auto/click crop saving, and easy deployment.

---

🚀 Features

✅ Detects multiple small & distant faces
✅ Real-time Auto & Click-to-Zoom modes
✅ Save cropped faces automatically to disk
✅ Works with webcam or video input
✅ Includes training visualizations and metrics
✅ Exportable to ONNX for deployment on edge devices

---

🎥 Demo (Before vs After)

• Left: Raw input video (4 sec)
• Right: Detected faces with bounding boxes (CerberusFace-YOLOv8)

---

🧠 Training Summary

Parameter	Value
Base Model	YOLOv8n (Nano)
Dataset	WIDER FACE
Epochs	200
Image Size	640x640
Batch Size	16
Optimizer	AdamW
Augmentations	Mosaic, MixUp, CopyPaste
Loss Weights	box: 7.5, cls: 0.5, dfl: 1.5

---

📊 Training Results

🔻 Loss & Metric Curves

📌 Highlights:

• Smooth decrease in box, cls, and DFL loss
• Precision: ~0.85, Recall: ~0.57
• mAP@0.5 ≈ 0.64, mAP@0.5:0.95 ≈ 0.36

---

🎨 Sample Training Batches

Example of how the model "saw" faces during training.

---

🎯 Validation Results

Detection results on validation samples — robust performance in dense/crowded scenes.

---

🛠️ Inference

Run face detection with:

python Scripts/Inference.py --source 0 --mode auto --save

Arguments:

• --source → Webcam index or video path
• --mode → auto (all faces) or click (select face with mouse)
• --save → Save cropped faces into face_data/

---

📂 Repository Structure

datasets/
  └── data.yaml
Model/
  ├── results 
     └──  cerberusface_yolov8n.pt
  ├── weights
    └──  confusion_matrix.png
    ├── results.png
    ├── train_batch.jpg
    └── val_batch_labels.jpg
Scripts/
  ├── Inference.py
  └── train.py
media/
  ├── before.gif             
  └── after.gif              
Requirements.txt
README.md
LICENSE

---

👥 Credits & Contributions

• Base Model: Ultralytics YOLOv8 (GPL-3.0 License)
• Dataset: WIDER FACE
• Project Author: Ajay Bharathi, Sridhar.S
• Organization: Cryptarchs
• Contributors: Open for community contributions!

📩 Contact: Open issues or PRs for suggestions and improvements.

---

📜 License

This project is released under the GPL-3.0 License.
You are free to use, modify, and redistribute it, provided that derivative works also remain open-source under GPL-3.0.

For full license terms, see the LICENSE file.

---

🌟 Future Work

🔹 Improve recall with larger models (YOLOv8m/s)
🔹 Add multi-class face attributes (e.g., mask/no-mask, emotion detection)
🔹 Optimize for mobile & edge deployment
🔹 Integrate with live CCTV pipelines

---

💡 How to Contribute

1. Fork the repository
2. Create your feature branch (git checkout -b feature/your-feature)
3. Commit your changes (git commit -m 'Add some feature')
4. Push to the branch (git push origin feature/your-feature)
5. Open a Pull Request

---

🔥 With CerberusFace-YOLOv8, you get a fast and accurate face detection model, fine-tuned for the real world.