SelectAnyTree: A Promptable Instance Segmentation Model for 3D Forest LiDAR Point Clouds

A new computer vision model called SelectAnyTree can isolate individual trees in 3D forest scans from a single user click, according to research posted to the arXiv preprint server on June 25, 2026 [1][2]. The model, proposed by researcher Trung Thanh Nguyen, is designed to work with LiDAR point clouds — dense three-dimensional data sets collected by laser scanning [1][2]. Forest monitoring increasingly relies on such scans for detailed measurement, but manually labeling individual trees within the data is slow and expensive. A single hectare can contain millions of points and hundreds of overlapping tree crowns, making annotation from scratch laborious and error-prone [2]. SelectAnyTree addresses this by adopting a promptable segmentation approach. A user clicks on a target tree, and the model encodes the click’s 3D position and polarity into a query that is decoded into a tree mask. The system also includes a Canopy Height Model-guided first prompt that automatically suggests treetops without any user input [2]. The decoder uses a state-space architecture to capture long-range context across large forest scenes with linear-time complexity [2]. In evaluations across seven forest regions and an independent held-out test set, SelectAnyTree segmented a target tree to 78.2 Intersection over Union from a single click, which the authors report is 24.8 points above the strongest promptable baseline [2]. The model also used fewer parameters and required less inference time than prior promptable models, according to the paper [2]. The work was posted on arXiv, an open-access repository that hosts preprints across physics, computer science, and other fields [6]. As of late 2024, arXiv was receiving roughly 24,000 new articles per month and had surpassed two million total submissions [6]. Papers on the platform are moderated but not peer-reviewed before posting [6]. The SelectAnyTree source code has been made available on GitHub [2]. Automated tree segmentation from LiDAR has become a focus area as governments and researchers seek scalable ways to track forest structure, carbon storage, and biodiversity. The promptable design of SelectAnyTree could reduce the annotation bottleneck that has slowed progress in the field [2].