BAMspecies: A Benchmark Dataset for Individual Tree Species Classification from Ultra High-Resolution RGB Imagery and LiDAR Point Cloud

In forestry individual tree species classification is particularly constrained by the need for expert annotations and high-resolution data capturing complex crown structures. While comprehensive datasets exist for general-purpose tasks, domain-specific applications requiring expert knowledge remain severely underrepresented. To address this gap, we present BAMspecies, a large-scale benchmark dataset for tree species classification in dense forests. BAMspecies contains approximately 25,000 manually annotated individual trees with expert-verified species labels, very high-resolution RGB orthomosaics with 2 cm GSD, and dense LiDAR point clouds with approximately 700 pts/m2 from unmanned aerial vehicles (UAV) data. The dataset provides species labels, instance annotations, and corresponding 2D and 3D representations, enabling standardized evaluation of tree species recognition from RGB imagery and point clouds. To establish reference performance, we evaluate seven representative 2D models and two 3D models, achieving mean F1-scores from 47.18% to 59.45% for RGB imagery and from 26.23 % to 53.32 % for point clouds. These results highlight the difficulty of fine-grained species recognition in structurally complex forests. By combining expert annotations, high-resolution RGB data, dense LiDAR, and reproducible benchmark splits, BAMspecies enables systematic evaluation of tree species classification methods across 2D imagery and 3D point clouds. The dataset is available on the BAMspecies website.