32,203 images containing 393,703 labeled faces categorized into 61 distinct event classes. The dataset features significant variability in facial scale, pose, and occlusion levels across its training, validation, and testing splits.
Use Cases
- Train face detection models to handle extreme scale and pose variations using the bounding box annotations
- Analyze model performance across different environmental contexts using the 61 event class labels
- Benchmark detection algorithms against the PASCAL VOC metric using the 50% testing set
Strengths
- 393,703 manually labeled face bounding boxes across 32,203 images
- Categorization into 61 event classes representing diverse real-world scenarios
- Standardized evaluation using the PASCAL VOC metric for detection accuracy
- Structured data splits of 40% training, 10% validation, and 50% testing