FaceIA v3.0: Installation Guide
1. Introduction
FaceIA automatically detects and tracks facial features (face, eyes, mouth, nose) and classifies facial expressions (emotions) in images and videos, generating Interest Area Set (IAS) files for EyeLink Data Viewer and Experiment Builder.
Key Improvements in Version 3.0:
2. Why Use FaceIA?
FaceIA is designed to eliminate the tedious, frame-by-frame manual labour historically required to map facial features in eye-tracking research. Researchers might use this tool to:
3. System Requirements & Installation
Windows 10 or 11:
macOS (M-Series):
4. Emotion Classification
FaceIA utilizes an advanced ONNX-based neural network to classify 7 primary facial expressions: Anger, Contempt, Disgust, Fear, Happiness, Sadness, and Surprise, alongside Neutral.
5. Troubleshooting
6. Credits & Attributions
This application relies on open-source libraries and advanced models to accomplish facial tracking and emotion classification. Sincere thanks to the developers and communities behind these technologies:
7. Support
For any issues not covered in this manual, please refer to support@sr-research.com.
1. Introduction
FaceIA automatically detects and tracks facial features (face, eyes, mouth, nose) and classifies facial expressions (emotions) in images and videos, generating Interest Area Set (IAS) files for EyeLink Data Viewer and Experiment Builder.
Key Improvements in Version 3.0:
- Emotion Tracking: Classifies 7 primary emotions + Neutral using an advanced neural network.
- Dedicated Emotion Export: Ability to emotionally classify a face and export a streamlined IAS containing only the facial boundary and its detected emotion.
- Combined IA & Emotion Export: Ability to export a full Interest Area set (including specific features like eyes, nose, and mouth) while appending the dominant emotion directly to the face feature labels.
- Live Emotion Preview: Visualize live emotion statistics and confidence scores directly on the media overlay before processing.
2. Why Use FaceIA?
FaceIA is designed to eliminate the tedious, frame-by-frame manual labour historically required to map facial features in eye-tracking research. Researchers might use this tool to:
- Automate Interest Area Generation: Instantly map the face, eyes, nose, and mouth across video frames or image batches.
- Analyse Emotional Responses: Track changes in a stimuli's facial expressions dynamically, quantifying exactly when and how strongly emotions like Happiness, Fear, or Surprise occur during presentation.
- Streamline Analysis: Export perfectly formatted `.ias` files that drop directly into EyeLink Data Viewer and/or Experiment Builder, ensuring stimulus dimensions and coordinates precisely match your experimental setup.
3. System Requirements & Installation
Windows 10 or 11:
- Download the installer: FaceIA_Setup.exe
- Click through the installer and once completed the application can be found in Start > All Apps > SR Research as FaceIA.
macOS (M-Series):
- Download the macOS installer package for Apple Silicon - coming soon
- Follow the standard installation prompts to move the application to your Applications folder.
4. Emotion Classification
FaceIA utilizes an advanced ONNX-based neural network to classify 7 primary facial expressions: Anger, Contempt, Disgust, Fear, Happiness, Sadness, and Surprise, alongside Neutral.
- Dynamic Vetoes: The system intelligently limits "Neutral" readings if a person's mouth is widely open (e.g., speaking or shouting).
- Adjustable Confidence Thresholds: You can independently adjust the required emotion confidence for Images and Videos within the ▶ Thresholds settings panel. If the model's certainty falls below your chosen threshold, it will safely default to Neutral or Unknown.
- Performance Tip: Emotion classification is mathematically intensive. The engine intelligently disables emotion processing when you are only saving standard Interest Areas, maximizing processing speed.
5. Troubleshooting
- No face detected: Lower the Min Detection Confidence or switch to Robust Mode. You can also manually draw the region using Add Face.
- Emotions say "Unknown": The emotion tracker uses a 5-frame rolling buffer to prevent flickering. If you switch tabs while paused, press Play or scrub the timeline forward a few frames to flush the buffer and reveal the emotions.
- Batch Processing speed: Robust Mode and Emotion Tracking are CPU intensive. For large batches without emotion requirements, use Standard Mode on the Interest Areas tab.
6. Credits & Attributions
This application relies on open-source libraries and advanced models to accomplish facial tracking and emotion classification. Sincere thanks to the developers and communities behind these technologies:
- MediaPipe Face Mesh (Google): The core technology used for accurate 3D facial landmark detection, enabling the precise identification of the face, eye, mouth, and nose boundaries. (More info: https://mediapipe.dev/)
- Emotion Classifier Model (enet_b0_8_best_vgaf.onnx): The facial emotion recognition capabilities are powered by the EfficientNet-B0 model (enet_b0_8_best_vgaf), a lightweight convolutional neural network optimized for affective behavior analysis. This model relies on the EmotiEffLib / HSEmotion library created by Andrey Savchenko and the HSE-NN team, demonstrating state-of-the-art accuracy trained across the AffectNet and VGAF (Video-level Group AFfect) datasets.
7. Support
For any issues not covered in this manual, please refer to support@sr-research.com.