There is no problem with the wider data set label. The WIDERFACE data set is a face detection benchmark data set. The pictures are selected from the WIDER Web Image Dataset for Event Recognition data set. The number of pictures is 32,203 and the number of faces is 393,703. Faces are highly transformed in different forms such as size, scale, position, pose, occlusion, etc.
Characteristics of wider dataset labels
WIDERFACE dataset is based on 61 event categories. Each event category is randomly selected for training, 40%, validation, 10% and testing. 50. Training and testing contain boundingbox groundtruth but validation does not. The detection algorithm is evaluated on the test set in the same way as PASCALVOCDATASET.
And the true bounding box of the test set has not been released. Participants can submit prediction files and the evaluation results will be given by WIDERFACE. The WiderPerson data set is a benchmark data set for pedestrian detection in crowded scenes. Images are selected from a variety of scenes no longer limited to traffic scenes.