WIDER Face

WIDER FACE 数据集是一个人脸检测基准数据集，其中图像选自公开可用的 WIDER 数据集。我们选择了 32,203 张图像并标记了 393,703 张在尺度、姿势和遮挡方面具有高度可变性的面孔，如样本图像中所示。 WIDER FACE 数据集基于 61 个事件类进行组织。对于每个事件类，我们随机选择 40%/10%/50% 的数据作为训练、验证和测试集。我们采用与 PASCAL VOC 数据集相同的评估指标。与 MALF 和 Caltech 数据集类似，我们不发布测试图像的边界框地面实况。用户需要提交最终的预测文件，我们将进行评估。

WIDER FACE is a benchmark dataset for face detection, with images sourced from the publicly available WIDER dataset. We selected 32,203 images and annotated 393,703 faces that exhibit high variability in scale, pose and occlusion, as demonstrated in the sample images. The WIDER FACE dataset is organized according to 61 event classes. For each event class, we randomly partition 40%/10%/50% of the data into training, validation and test sets, respectively. We employ the same evaluation metrics as those utilized in the PASCAL VOC dataset. Similar to the MALF and Caltech datasets, we do not release the ground-truth bounding boxes for the test images. Users are required to submit their final prediction files for evaluation.