v1.3.1

What's Changed

• fix: Fix type conversion and remove redundant type conversion by @yakhyo in #29

Full Changelog: v1.3.0...v1.3.1

v1.3.0

What's Changed

• feat: Update examples and some minor changes to UniFace API by @yakhyo in #28

Full Changelog: v1.2.0...v1.3.0

v1.2.0

What's Changed

• feat: Add YOLOv5 face detection support by @yakhyo in #26

Full Changelog: v1.1.2...v1.2.0

v1.1.2

What's Changed

• feat: Some minor changes to code style and warning supression by @yakhyo in #23
• feat: Enhace emotion inference speed on ARM and add FaceAnalyzer, Face classes for ease of use. by @yakhyo in #25

Full Changelog: v1.1.1...v1.1.2

v1.1.1

Full Changelog: v1.1.0...v1.1.1

v1.1.0

What's Changed

• ref: Add comprehensive test suite and enhance model functionality by @yakhyo in #18

Full Changelog: v0.1.9...v1.1.0

Model Weights

UniFace: A Comprehensive Library for Face Detection, Recognition, Landmark Analysis, Age, and Gender Detection

Features

• High-speed face detection using ONNX models (Added: 2024-11-20)
• Accurate facial landmark localization (e.g., eyes, nose, and mouth) (Added: 2024-11-20)
• Face recognition with multiple model options (Added: 2024-11-20)
• Age and gender detection (Added: 2024-11-20)
• Emotion recognition (Added: 2024-11-20)
• Easy-to-use API for inference and visualization (Added: 2024-11-20)

Available Models

Face Detection Models

RetinaFace Models

Model Name	Easy	Medium	Hard	Use Case
retinaface_mnet025	88.48%	87.02%	80.61%	Mobile/Edge devices
retinaface_mnet050	89.42%	87.97%	82.40%	Mobile/Edge devices
retinaface_mnet_v1	90.59%	89.14%	84.13%	Balanced mobile
retinaface_mnet_v2	91.70%	91.03%	86.60%	Recommended default
retinaface_r18	92.50%	91.02%	86.63%	Server/High accuracy
retinaface_r34	94.16%	93.12%	88.90%	Maximum accuracy

SCRFD Models

Model Name	Easy	Medium	Hard	Use Case
scrfd_500m	90.57%	88.12%	68.51%	Real-time applications
scrfd_10g	95.16%	93.87%	83.05%	High accuracy + speed

Face Recognition Models

ArcFace

• arcface_mnet: MobileNet backbone - Balanced performance (recommended)
• arcface_resnet: ResNet50 backbone - Maximum accuracy

Trained on MS1M-V2 dataset (5.8M images, 85K identities)

MobileFace

Model Name	Backbone	LFW	CALFW	CPLFW	AgeDB-30	Use Case
mobilenetv1_025	MobileNetV1 0.25	98.76%	92.02%	82.37%	90.02%	Ultra-lightweight
mobilenetv2	MobileNetV2	99.55%	94.87%	86.89%	95.16%	Mobile/Edge
mobilenetv3_small	MobileNetV3-S	99.30%	93.77%	85.29%	92.79%	Mobile optimized
mobilenetv3_large	MobileNetV3-L	99.53%	94.56%	86.79%	95.13%	Balanced mobile

Trained on MS1M-V2 dataset (5.8M images, 85K identities)

SphereFace

Model Name	Backbone	LFW	CALFW	CPLFW	AgeDB-30	Use Case
sphere20	Sphere20	99.67%	95.61%	88.75%	96.58%	Research/Comparison
sphere36	Sphere36	99.72%	95.64%	89.92%	96.83%	Research/Comparison

Trained on MS1M-V2 dataset (5.8M images, 85K identities)

Facial Analysis Models

Landmark Detection

• 2d_106: 106-point facial landmark detection

Age & Gender Detection

• age_gender: Predicts age and gender from face regions

Emotion Recognition

• affecnet7: 7-class emotion recognition (Neutral, Happy, Sad, Surprise, Fear, Disgust, Anger)
• affecnet8: 8-class emotion recognition (adds Contempt)

Installation

pip install uniface

For GPU acceleration:

pip install uniface[gpu]

Model Selection Guide

By Use Case

• Mobile/Edge: RetinaFace MNET_025 + MobileFace MNET_V2
• Real-time: SCRFD 500M + ArcFace MNET
• High Accuracy: SCRFD 10G + ArcFace RESNET
• Server/Cloud: RetinaFace R34 + ArcFace RESNET

By Hardware

• Apple Silicon: All models work well with ARM64 optimizations
• NVIDIA GPU: Larger models for maximum throughput
• CPU Only: Lightweight models recommended

Available Scripts

• run_detection.py: Face detection with different models
• run_recognition.py: Face recognition and similarity comparison
• run_landmarks.py: Facial landmark detection
• run_age_gender.py: Age and gender prediction
• run_video_detection.py: Video processing
• run_face_search.py: Face search system
• batch_process.py: Batch processing capabilities
• download_model.py: Model management

Benchmarking

Test performance on your hardware:

python scripts/run_detection.py --iterations 100

Documentation

• QUICKSTART.md - Getting started guide
• MODELS.md - Complete model documentation
• Examples - Jupyter notebooks with examples

v1.0.0

Full Changelog: v0.1.2...v1.0.0

🎉 UniFace v1.0.0 - First Stable Release

This is the first production-ready release of UniFace with a complete feature set, stable API, and comprehensive documentation.

✨ What's New

High-Level Pipeline API

• FacePipeline - Comprehensive face analysis workflows
• process_faces() - Task-based processing (detect, landmarks, recognize, age_gender, emotion)
• compare_faces() - Compare faces between two images
• extract_face_embeddings() - Extract embeddings for face recognition
• analyze_face_attributes() - Analyze age, gender, and emotion

Face Detection

• RetinaFace detector (MobileNet, ResNet variants)
• SCRFD detector (10G, 500M variants)
• Unified detection output format

Face Recognition

• ArcFace models (MobileNet, ResNet)
• MobileFace models (lightweight)
• SphereFace models (angular margin)
• Normalized embedding extraction

Landmark Detection

• 106-point facial landmark detection
• Landmark-based face alignment

Attribute Analysis

• Age and Gender Detection
• Emotion Recognition (7 categories)

Documentation & Examples

• Comprehensive README with quick start
• examples/README.md with detailed usage guide
• examples/basic_usage.py - 6 practical examples
• examples/advanced_usage.py - Advanced patterns (batch processing, video, face database)
• Jupyter notebooks for tutorials

🚀 Quick Start

import uniface

# Simple face detection
faces = uniface.detect_faces(image, method='retinaface')

# Comprehensive analysis
faces = uniface.process_faces(
    image,
    tasks=['detect', 'landmarks', 'age_gender', 'emotion']
)

# Face comparison
result = uniface.compare_faces(image1, image2, threshold=0.6)

📦 Installation

pip install uniface

🔧 What Changed

• API Standardization: Unified data formats across all modules
• Factory Functions: create_detector(), create_recognizer(), create_landmarker(), create_attribute_predictor()
• Module Organization: Clean separation of detection, recognition, landmark, and attribute modules
• Documentation: Simplified to maintainable Markdown files
• Apple Silicon Support: Added onnxruntime-silicon support
• Improved Logging: Verbose flag for better debugging

🗑️ Removed

• Legacy RetinaFace implementation (consolidated into detection module)
• Complex MkDocs documentation setup
• Duplicate code

🐛 Fixed

• API format inconsistencies
• Code duplication issues
• Import structure and module exports
• Installation issues with onnxruntime-gpu

📚 Documentation

• README
• Examples
• PyPI Package

---

Full Changelog: https://github.com/yakhyo/uniface/commits/v1.0.0

Full Changelog: v0.1.9...v1.0.0