COFW(Caltech Occluded Faces in the Wild)

Caltech Occluded Faces in the Wild (COFW) 数据集旨在呈现真实世界条件下的人脸。由于姿势、表情、太阳镜和帽子等配件的使用以及与物体（例如食物、手、麦克风等）的交互作用的差异，人脸的形状和遮挡有很大差异。所有图像均使用与 LFPW 中相同的 29 个地标进行手工注释。地标位置以及它们的遮挡/未遮挡状态都被注释了。面部被不同程度地遮挡，遇到的遮挡类型差异很大。 COFW 的平均遮挡超过 23。

Caltech Occluded Faces in the Wild (COFW) dataset is designed to showcase human faces in real-world scenarios. Faces undergo substantial variations in both shape and occlusion status, caused by factors including diverse poses, expressions, accessories like sunglasses and hats, as well as interactions with various objects (e.g., food, hands, microphones, and so on). All images are manually annotated with 29 facial landmarks identical to the set used in LFPW. Both the precise coordinates of these landmarks and their occlusion/non-occlusion states are annotated. Faces in the dataset are occluded to varying degrees, with a wide range of occlusion types encountered. The average occlusion level of the COFW dataset exceeds 23.

搜集汇总

背景与挑战

背景概述

COFW数据集是一个专注于真实世界遮挡人脸的图像数据集，由加州理工学院于2013年发布。它包含手工标注的29个人脸地标位置及其遮挡状态，平均遮挡率超过23%，旨在支持人脸对齐和遮挡鲁棒性研究。