SHREC 2021: 3D point cloud change detection for street scenes

The rapid development of 3D acquisition devices enables us to collect billions points in a few hours. However, the analysis output data is challenging task, especially field point cloud change detection. In this Shape Retrieval Challenge (SHREC) track, we provide street-scene dataset for consists 866 object pairs year 2016 and 2020 from 78 large-scale street scene clouds. Our goal detect changes multi-temporal clouds complex environment. We compare three methods on benchmark, with one handcrafted (PoChaDeHH) other two learning-based (HGI-CD SiamGCN). results show that algorithm has balanced performance over all classes, while achieve overwhelming but suffer class-imbalanced problem may fail minority classes. randomized oversampling metric applied SiamGCN can alleviate problem. Also, different siamese network architecture HGI-CD contribute designing detection task.