LidarDome: A Multi-View LiDAR Scanner 3D Point-Cloud Dataset for Heritage Apple Phenotyping

LidarDome 2023 is a curated multi-view 3D point-cloud dataset of detached apples acquired during the 2023 harvest from the Swiss Apple Core Collection (SACC). The dataset comprises 33 heritage apple cultivars, with approximately 24 fruits per cultivar sampled from three trees. In total, the release contains 7,910 cleaned single-camera point clouds. The data were acquired using LidarDome, a static scanner consisting of ten Intel RealSense L515 LiDAR cameras arranged to capture overlapping depth and color views around the fruit. The system records a complete multi-view scan in approximately 45 seconds without mechanically rotating either the specimen or the sensors. The dataset contains: - cleaned RGB-colored point clouds from the individual camera positions, C0–C9, before final multi-view stitching;- calibration-set-specific 4 × 4 transformation matrices for registering the individual views;- final stitched and stalk-removed colored point clouds of the fruit body;- a mapping file linking tree identifiers to cultivar labels. The 2023 dataset was generated using a manually verified Pose Graph Optimization pipeline. The procedure combines an initial camera alignment, manually selected correspondences between overlapping views, point-to-point ICP refinement, and global pose-graph optimization. The automated sphere-target calibration evaluated in the associated publication was developed subsequently and was not used to generate this dataset. Preprocessing includes removal of the background and blue fruit supports, statistical outlier filtering, removal of stalk-like structures, transformation and merging of the individual camera views, 1 mm voxel-grid downsampling, and final point-cloud cleaning. The transformation matrices are specific to their respective calibration sets. Calibration sets 2 and 3 contain distinct transformation matrices because one camera was exchanged during the 2023 scanning campaign. The dataset is intended to support research on multi-view point-cloud registration, fine-grained apple cultivar classification, geometric and appearance feature extraction, and pomology-inspired morphometric analysis. The individual camera point clouds represent partial views of each fruit, whereas the final stitched point clouds combine the available views into a multi-view representation of the fruit body. Associated publication: Wróbel, A. et al. “LidarDome: A Multi-View LiDAR Scanner and 3D Point-Cloud Dataset for Heritage Apple Phenotyping.”