About LunaNet AI
The Future That Grows With You
Our Technology
LunaNet AI represents the cutting edge of facial expression recognition technology, powered by the largest visual cross-corpus study in emotion recognition to date.
Advanced Architecture
Our system employs a sophisticated two-stage architecture:
- Backbone Model: Based on VGGFace2 ResNet50, fine-tuned on the AffectNet dataset with 66.4% accuracy
- Temporal Processing: LSTM-based models trained on dynamic datasets for robust temporal analysis
Cross-Corpus Training
Our models are trained on multiple diverse datasets including:
- RAVDESS (Dynamic expressions)
- CREMA-D (Crowd-sourced emotions)
- SAVEE (Audio-visual emotions)
- RAMAS (Russian multimodal corpus)
- IEMOCAP (Interactive emotional capture)
- Aff-Wild2 (In-the-wild expressions)
Performance Metrics
66.4%
AffectNet Accuracy
7
Emotion Classes
Real-time
Processing
Multiple
Dataset Support
Cross-Corpus Results
Our leave-one-corpus-out cross-validation demonstrates robust performance across diverse datasets:
| Test Dataset | UAR (%) | Performance |
|---|---|---|
| Aff-Wild2 | 51.6% | Good |
| RAVDESS | 65.8% | Excellent |
| CREMA-D | 60.6% | Excellent |
| SAVEE | 76.1% | Outstanding |
| RAMAS | 44.3% | Fair |
| IEMOCAP | 25.1% | Limited |
Key Features
- Real-time Processing: Live emotion detection via webcam
- Multi-format Support: Images and videos
- High Accuracy: State-of-the-art performance
- Robust Detection: Works in various conditions
- Detailed Analysis: Comprehensive emotion reports
Technical Specifications
- Framework: TensorFlow & PyTorch
- Face Detection: RetinaFace
- Backbone: VGGFace2 ResNet50
- Temporal: LSTM Networks
- Input Size: 224x224 pixels
- Processing: GPU accelerated
Use Cases
- Educational Research
- Mental Health Monitoring
- Customer Experience
- Gaming & Entertainment
- Security & Surveillance
Get Started Today
Experience the power of advanced emotion detection technology