32,203 images containing 393,703 labeled face instances across various scales, poses, and occlusion levels. These annotations are provided in the Pascal VOC XML format, facilitating integration with object detection frameworks like Faster R-CNN or SSD.
Use Cases
- Train face detection models using the <bndbox> coordinates and <name> labels within the XML files.
- Evaluate model performance on the 'face' class using the standardized XML ground truth.
- Integrate WIDER FACE data into frameworks that require Pascal VOC XML inputs for object detection.
Strengths
- Annotations provided in standard Pascal VOC XML format including <object> and <bndbox> tags.
- Includes bounding box coordinates (xmin, ymin, xmax, ymax) for every detected face.
- Covers 32,203 images with high variability in face scale and orientation.
- Includes <name> tags identifying the 'face' class for each annotation.