Large-Scale Colored Point Cloud Upsampling Dataset

In order to develop and analyse the performance of large-scale colored point cloud upsampling, we built a large-scale colored point cloud dataset for training and evaluating the upsampling network. This large-scale colored point cloud dataset consists of 121 original colored point clouds, 43 of which were scanned by us, while the other 78 were obtained from the SIAT-PCQD, Moving Picture Experts Group (MPEG) point cloud, and Greyc 3D colored mesh database. These point clouds cover six categories, including animals, plants, toys, sculptures, people and others. Most of the point clouds are distributed between 80,000 and 1,500,000 points, with the largest point cloud containing 3,817,422 points and the smallest point cloud containing 19,247 points. Compared to existing sparse point cloud datasets, our point clouds are larger in size and have a more diverse range of distributions. By calculating depth complexity and texture complexity, we also demonstrate the diversity of these point clouds in terms of depth and texture complexity.