davideuler / HRNet-Landmarks

Web application for Realtime landmark for Video frames

🎯 High-Resolution Facial Landmark Detection

State-of-the-art 68-point facial landmark detection using High-Resolution Networks (HRNet).

---

✨ Features

• 🚀 High-Resolution Networks: Maintains high resolution throughout the process
• 🎯 68 Facial Landmarks: Precise detection of key facial points
• ⚡ Real-time Performance: Optimized for live video processing
• 🔥 Multi-Scale Fusion: Advanced feature fusion across different resolutions
• 📊 Low NME Score: Superior accuracy on 300W dataset

---

Usage

Datasets: 300W

• Download the datasets from official sources.
• Download the annotations files from OneDrive.

📈 Training

To train the model, run:

• Configure your dataset path in utils/config.py for training

$ python main.py --train

Testing

For testing the model, use:

$ python main.py --test

Real-Time Demo

To run the real-time facial landmark detection:

$ python main.py --demo

📊 Performance: 300W

NME	NME	pretrained model	model
HRNetV2-W18	3.3	ImageNet	best.pt

Reference

https://github.com/HRNet/HRNet-Facial-Landmark-Detection