py-feat-application

Facial expression analysis application using Py-Feat to extract Action Units (AUs) and emotions from videos.

Features

• Extract 20 Action Units from video frames
• Extract 7 emotions (anger, disgust, fear, happiness, sadness, surprise, neutral)
• Process videos at 30fps (configurable)
• Supports videos of any length - from seconds to hours
• Output to CSV and NumPy formats
• Comprehensive analysis and visualization tools

Quick Start

Extract AUs and Emotions from Video

python extract_au_emotions.py your_video.mp4

The script automatically handles videos of any duration.

Example output for a 6-minute video at 30fps:

• AU data: (10,800 frames × 20 AUs)
• Emotion data: (10,800 frames × 7 emotions)
• Combined data in both CSV and NPY formats

Processing time: ~1-2 seconds per frame on CPU

Installation

# Activate the virtual environment
source venv/bin/activate

# Or use the activation script
source activate_pyfeat.sh

Scripts

• extract_au_emotions.py - Main script to extract AUs and emotions from video
• detect_video.py - Original video detection script
• analyze_results.py - Analysis script for results
• create_annotated_video.py - Create annotated video with detections

Documentation

• USAGE_GUIDE.md - Detailed usage guide and examples
• QUICKSTART.md - Quick start guide

Output Format

NumPy Arrays (.npy)

• *_au_data.npy - Shape: (num_frames, 20) - AU values
• *_emotion_data.npy - Shape: (num_frames, 7) - Emotion values
• *_au_emotions.npy - Shape: (num_frames, 27) - Combined data

CSV Files

• *_au_emotions.csv - Compact CSV with frame, time, AUs, and emotions
• *_full_predictions.csv - Complete predictions with all columns

References

• Py-Feat - Python Facial Expression Analysis Toolbox
• FACS - Facial Action Coding System