DNA-Rendering is a substantial, high-quality human performance data repository developed for enhancing neural actor rendering. The work aims to overcome the limitations in existing datasets and to cater to various real-world scenarios requiring robust methods. The repository comprises over 1500 human subjects, 5000 motion sequences, and 67.5M frames, ensuring an extensive range of geometry and appearance variations. It provides various assets to each subject, such as 2D/3D human body keypoints, SMPLX models, cloth/accessory materials, multi-view images, and videos. These resources significantly improve the accuracy of current rendering tasks. A professional multi-view system captures the data, guaranteeing top-tier resources for training and evaluation.

 

Publication date: 19 Jul 2023
Project Page: https://dna-rendering.github.io/
Paper: https://arxiv.org/pdf/2307.10173.pdf