CVPR 2022 | clonedperson: building a large-scale virtual pedestrian data set of real wear and wear from a single photo

In this paper, we clone a whole set of clothes from a single photo to wear three-dimensional characters , Build a system that contains 5621 A virtual pedestrian data set of three-dimensional human models ClonedPerson. These virtual characters simulate real monitoring in the game environment and render multi camera videos in multiple scenes . Experiments show that , This data set has achieved good generalization in the task of pedestrian re recognition , And can be applied to unsupervised domain adaptation 、 Unsupervised learning 、 Human key point detection and other related tasks . Related papers have been published by CVPR 2022 receive , Data and code are open source .

• Address of thesis ：https://arxiv.org/pdf/2204.02611.pdf
• Project address ：https://github.com/Yanan-Wang-cs/ClonedPerson