WFLW (Wider Facial Landmarks in the Wild)

Wider Facial Landmarks in-the-wild (WFLW) 包含 10000 个人脸（7500 个用于训练，2500 个用于测试）和 98 个完全手动注释的地标。除了地标标注外，新的数据集还包括丰富的属性标注，即遮挡、姿势、化妆、照明、模糊和表达，用于对现有算法的综合分析。与之前的数据集相比，所提出的数据集中的人脸在表情、姿势和遮挡方面引入了很大的变化。

Wider Facial Landmarks in-the-wild (WFLW) consists of 10,000 facial images, with 7,500 samples allocated for training and 2,500 for testing, all annotated with 98 fully manually labeled facial landmarks. In addition to landmark annotations, this novel dataset also provides comprehensive attribute annotations covering occlusion, pose, makeup, illumination, blur, and expression, which are designed for comprehensive analysis of existing algorithms. Compared with prior facial landmark datasets, the faces in the proposed WFLW dataset exhibit substantial variations in terms of expression, pose, and occlusion conditions.

搜集汇总

背景与挑战

背景概述

WFLW数据集包含10,000个人脸图像，其中7,500个用于训练、2,500个用于测试，并提供了98个完全手动注释的地标。此外，数据集还包括遮挡、姿势、化妆等多种属性标注，用于对算法进行综合分析。相较于以往数据集，WFLW在表情、姿势和遮挡方面引入了更大的变化。