PersonHD: A High-Difinition and Large-Scale Person Image Dataset Dataset release : A high-definition person image dataset that contains 4000 videos (4000 =100id20 Action2 view ) with 299K frames,which collected in 2019 for the Pose-guided person generation task. The data can be obtained by sending an email to me (tongxueqing2020@ia.ac.cn) via an official email (not use qq, gmail, etc.) including your institution/company information and the purpose on the dataset.
There are 100 subjects in PersonHD dataset. The same subject has many pictures of different poses with higher fine-grained actions, more actions (20 types in total, see action descriptions for details), and poses that are closer to the daily life of the human body , the picture resolution is 1520*880, higher definition, so it has more detailed ID facial features. The single person is unobstructed, the light is good, and the background is two-fold, white cloth and daily scene, which is closer to real life. We provide three modes：img/seg(mask)/pose. Segmentation is generated by open-source JPPNet model (https://github.com/Engineering-Course/LIP_JPPNet). Human skeleton keypoint is generated by open-source Openpose model.