ForestSemantic: A Dataset for Forest Semantic Learning from Close-Range Sensing Point Clouds

The ForestSemantic dataset provides detailed annotation of typical objects in forest scenes such as grounds, under-growths, trees, and different parts of a tree, such as stem, branches, and foliage.

The dataset supports studies at both individual tree- and plot-scales and to a branch level of detail, such as the detection, segmentation, and reconstruction of individual trees and branches. Particularly, this dataset supports detailed branch-level studies, as the branch annotations at an individual tree level in sample plots is rare, and is crucial for branch level studies. The dataset is expected to facilitate research in new dimensions, development of new method, and benchmarks of different systems and solutions.

Fig.1 The semantic labels of ground and non-ground points

Fig.2 The plot-level semantic labels of the Trunk, First-order branch, Higher-order branch and Foliage. (a)-(c) Trunk. (d)-(f) First-order branch. (g)-(i) Higher-order branch. (j)-(l) Foliage.

Fig.3 The semantic labeling of an individual tree.

Fig.4 The instance labeling of individual tree. (a)-(c) Top view. (d)-(f) Front view. (g)-(i) Details.

Fig.5 The instance labeling of individual First-order branch.

This dataset also provides rich measurements on tree structural traits that can be used as reference information for tree trait estimation methods. For each individual tree, the measured structure traits include the tree height, the diameter at the breast height (DBH), the crown projection area, the crown surface area, the crown volume, and the height of the first living branch. For each individual first-order branch of a tree, the positions of the starting and ending points, and the branch length are provided, which can be easily used to calculate further metrics such as the branch angle.

Fig.6 The reference of tree structure traits. (a) DBH. (b) Height. (c) Crown projection area. (d) Height of first-living branch. (e) Crown surface area and volume. (f) Length of First-order branch and its corresponding start and end points.

ForestSemantic is an open forest semantic dataset for non-commercial use under the license CC BY-NC-ND 4.0.

ForestSemantic is available at https://zenodo.org/records/13285640.

Plot 1 is available now and other plots will be published soon.