Point Cloud Understanding
Overview
Point cloud understanding studies how to interpret unordered 3D measurements for tasks such as object classification, scene segmentation, shape analysis, registration, and motion recognition. Methods range from local geometric descriptors and graph or point-transformer networks to voxel and bird’s-eye-view models, with newer work using self-supervised learning, multimodal language models, and parameter-efficient adaptation. Benchmarks such as ScanNet, S3DIS, SemanticKITTI, ShapeNetPart, and ScanObjectNN test performance across indoor, outdoor, and object-level settings. A central challenge is building models that preserve fine geometric detail while generalizing across sensors and domains, especially when labels are scarce or point sampling is uneven.
Latest papers
45 papers5 papers in the last four weeks, up 67% on the four weeks before. 0.0% of all new papers.