"Handheld SLAM\u2013LiDAR Stem Modeling: A Benchmark and Robust Pipeline"

"This dataset provides a fully annotated benchmark for handheld SLAM\u2013LiDAR forest stem modeling in dense Pinus radiata plantations. The benchmark targets practical failure modes that frequently bias diameter\/height estimation in operational surveys, including sparse or locally missing ground returns, trajectory-driven non-uniform sampling density, and branch\/understory contamination that yields partial and cluttered stem cross-sections. The release contains 40 fixed-area circular plots (radius 11.28 m) collected in two plantations in Western Australia using a handheld SLAM\u2013LiDAR system (Hovermap-class, VLP-16). For each plot, we provide (i) a registered 3D point cloud in a plot-centric coordinate frame, (ii) a terrain-normalized point cloud and the fitted ground surface grid for reproducible height normalization, and (iii) per-point stem instance annotations for all visible stems within the plot boundary. Ground-truth field measurements are provided for all standing trees, including DBH (1.3 m) and total height. To support stem-form validation beyond DBH, three plots include a destructively sampled subset with detailed diameter\u2013height profiles measured along the stem (base, 0.65 m, 1.30 m, 2.0 m, and subsequent intervals). The dataset additionally includes fixed benchmark splits and standardized evaluation tables to enable fair comparisons across stem extraction and geometric reconstruction pipelines. This benchmark is designed for research on robust stem segmentation, axis tracking, taper modeling, and inventory-grade metric estimation from handheld mobile LiDAR."