SoF(Specs on Faces)

Specs on Faces (SoF) 数据集，包含 42,592 (2,662×16) 张图像，用于 112 名在不同照明条件下戴眼镜的人（66 名男性和 46 名女性）。该数据集是免费的，可用于合理的学术公平使用。该数据集对人脸检测和识别提出了新的挑战。它专注于两个挑战：恶劣的照明环境和人脸遮挡，这对人脸检测、识别和分类有很大影响。眼镜是数据集所有图像中常见的自然遮挡。但是，每个图像都添加了另外两个合成遮挡（鼻子和嘴巴）。此外，可能会避开面部检测器和面部识别系统的三个图像过滤器被应用于每个图像。所有生成的图像都分为三个难度级别（简单、中等和困难）。这将图像数量扩大到 42,592 张图像（26,112 张男性图像和 16,480 张女性图像）。每张图片都有元数据，其中包含许多信息，例如：主题 ID、面部标志、面部和眼镜矩形、性别和年龄标签、照片拍摄年份、面部表情、眼镜类型等。

The Specs on Faces (SoF) dataset contains 42,592 (2,662×16) images, focusing on 112 individuals wearing glasses under varying lighting conditions, with 66 males and 46 females among them. This dataset is free and available for fair academic use.

It poses novel challenges for face detection and recognition tasks, with two core targeted challenges: harsh lighting environments and face occlusions, which significantly impact face detection, recognition and classification. Glasses serve as common natural occlusions across all images in the dataset. Additionally, two extra synthetic occlusions (nose and mouth) are added to each image. Furthermore, three image filters that can evade facial detectors and facial recognition systems are applied to every image.

All generated images are categorized into three difficulty levels: easy, medium and hard, which expands the total number of images to 42,592, consisting of 26,112 male images and 16,480 female images. Each image is accompanied by metadata containing various information, such as subject ID, facial landmarks, facial and glasses bounding boxes, gender and age labels, photo capture year, facial expressions, glasses types, and more.

搜集汇总

背景与挑战

背景概述

SoF数据集是一个包含42,592张图像的人脸识别数据集，专注于恶劣照明和人脸遮挡的挑战，图像分为三个难度级别，并附有详细的元数据信息。