FaceOcc

FaceOcc是一个大规模、高质量的人脸遮挡数据集，由里昂中央理工学院数学与信息科学系LIRIS实验室创建。该数据集包含超过30,000张高分辨率的人脸图像，涵盖多种遮挡类型，如太阳镜、眼镜、手、口罩等。创建过程中，研究人员对CelebA-HQ数据集中的图像进行了详细的遮挡标注，并通过网络收集了额外的遮挡类型。FaceOcc数据集的应用领域广泛，主要用于人脸相关任务，如人脸对齐、3D重建和人脸识别，旨在提高算法在复杂遮挡情况下的准确性和鲁棒性。FaceOcc is a large-scale, high-quality face occlusion dataset developed by the LIRIS Laboratory, Department of Mathematics and Information Science, École Centrale de Lyon. This dataset contains over 30,000 high-resolution face images covering diverse occlusion types, including sunglasses, eyeglasses, hands, face masks, and others. During its construction, researchers conducted detailed occlusion annotations on images from the CelebA-HQ dataset, and collected additional occlusion types from online sources. FaceOcc has broad application prospects, primarily being utilized for face-related tasks such as face alignment, 3D reconstruction, and face recognition, with the goal of enhancing the accuracy and robustness of algorithms in scenarios with complex occlusions.

搜集汇总

背景与挑战

背景概述

FaceOcc是一个大规模、高质量的人脸遮挡数据集，包含超过30,000张高分辨率图像，涵盖太阳镜、眼镜、手、口罩等多种遮挡类型。它基于CelebA-HQ数据集进行详细标注并补充额外数据，主要用于人脸对齐、3D重建和人脸识别任务，旨在提高算法在复杂遮挡情况下的准确性和鲁棒性。