LaPa
收藏资源简介:
LaPa数据集由京东AI研究院创建,包含22,000张具有表情、姿态、遮挡等多样性变化的人脸图像。每张图像均通过半自动标注框架进行精确的11类像素级标签标注,以及106个点的地标坐标。该数据集是目前最大的人脸解析公共数据集,旨在推动人脸解析技术的发展,特别是在面部美化、图像合成等领域的应用。创建过程中,通过高效的密集地标标注和像素级解析标注两个模块,显著简化和加速了标注过程。
The LaPa dataset was created by JD AI Research, consisting of 22,000 facial images with diverse variations in expression, pose, occlusion and other aspects. Each image is accurately annotated with 11 categories of pixel-level labels and 106-point landmark coordinates via a semi-automatic annotation framework. As the largest public facial parsing dataset to date, it aims to promote the development of facial parsing technologies, especially their applications in fields such as facial beautification and image synthesis. During its creation, two modules, namely efficient dense landmark annotation and pixel-level parsing annotation, were adopted to significantly simplify and accelerate the annotation workflow.



