Towards Better Point Cloud Visualizations for Tree Segmentations

Abstract

Instance segmentation of 3D forest point clouds is essential for applications such as forest monitoring, but dense canopy structure, heavy occlusion, and limited ground truth make quality assessment difficult. In practice, experts still need to visually inspect automated tree segmentations, yet conventional 3D views do not adequately support this task. This poster presents a design space for visualizing individual tree segmentations, organized around the key challenges of reducing occlusion, exposing relationships between segmented trees and neighboring points, and preserving global forest context. Our goal is to better support qualitative evaluation of segmentation quality in real-world forest point cloud data.