3DLF-Scan dataset: Multi-sensor 3D light-field and structured-light scans of 3D-printed Stanford figures.

Abstract

This data article presents 3DLF-Scan, a multi-sensor dataset of 3D-printed canonical Stanford models (bunny, dragon, asian_dragon, armadillo, happy, lucy, thai_statue) captured under controlled tabletop conditions. Each object is recorded in a full 360° turntable sweep with two calibrated light-field cameras (apiCAM PRO and apiCAM CUBE, photonicSENS) and a structured-light Revopoint Miraco 3D scanner. For every light-field viewpoint, the dataset includes RGB images, raw/sparse depth, dense depth completion in metric units, per-view foreground masks, and metric point clouds. Camera-from-object poses are provided as 4 × 4 matrices obtained from a nominal 5° turntable step refined by depth-only ICP under a single-axis constraint. For each figure, a separate scanner-based reference reconstruction and per-sensor calibration files (intrinsics, undistortion maps, checkerboard images) are also included. All assets are organized per object and per modality using standard formats (PNG, NPY, PLY, JSON) and naming conventions compatible with common 3D vision toolchains. The dataset is intended for developing and benchmarking methods in multi-view 3D reconstruction, depth completion, volumetric fusion, and cross-sensor registration.