TLS forest instance segmentation benchmark: 2983 manually segmented trees from four plots

This dataset was used for the analysis of the following publication:

Wout Cherlet, Karun Dayal, Shilin Chen, Zane Cooper, Mathias Disney, Andreas Hanzl, Shaun Levick, Joanne Nightingale, Niall Origo, Cornelius Senf, Luna Soenens, Louise Terryn, Wouter A.J. Van den Broeck, Kim Calders, Benchmarking tree instance segmentation of terrestrial laser scanning point clouds, ISPRS Journal of Photogrammetry and Remote Sensing, Volume 231, 2026, Pages 230-247, ISSN 0924-2716, https://doi.org/10.1016/j.isprsjprs.2025.10.033.

Any use of this dataset should cite the paper above (Creative Commons Attribution 4.0 International Public License).

Contact: wout.cherlet@ugent.be

----------------------------------------------------------------------------------------------

General information:

All data was collected using similar RIEGL VZ-series terrestrial laser scanners, and processed using Riscan Pro. The four plots are located at Litchfield, Australia; Wytham Woods, UK; Ofental, Germany; and Robson Creek, Australia, and are classified as tropical savanna, deciduous temperate forest, coniferous temperate forest and tropical rainforest respectively.

Trees were manually extracted by a total of nine operators. A second segmentation iteration was performed by a single operator for quality control and to ensure consistent level of detail. 

For additional details on data collection and processing, please refer to the publication above.

File format:

Each plot is divided in labeled plot point clouds for the test, validation and training areas. All files are in ply format, with following scalar fields: semantic (0: ground, 1: tree) and instance (sequential, starting at 1, -1 for ground points). They may be merged to form a single, labeled plot pointcloud.

Additionally, individual tree point clouds are attached in a zip container, divided between test, validation and training folders. Please note that all trees overlapping with each area are in the corresponding folders, so some edge trees are duplicated in multiple folders.

Evaluation code:

The evaluation workflow used in the publication above, can be replicated on this data using the code available at https://github.com/qforestlab.

Weights:

Finetuned weights for both TreeLearn and ForAINet (2 class semantic head) are available.

----------------------------------------------------------------------------------------------

Visualisations:

Left: top-down view of the plot point cloud, divided into testing, validation and training areas. Right: a lateral view of a subsection of the point cloud. The red box labeled A corresponds to the lateral view subsection. Points are colored by individual tree instance.

----------------------------------------------------------------------------------------------

Funding:

WC, ZC, WVDB and KC are funded by the European Union (ERC-2021-STG Grant agreement No. 101039795). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them. KD is funded by the Fonds Wetenschappelijk Onderzoek (FWO) - Project number G096223N, and Deutsche Forschungsgemeinschaft (DFG) - Project number 509915426. CS and AH received money from the Deutsche Forschungsgemeinschaft (DFG) - Project number 509915426. Shilin Chen is supported by the China Scholarship Council (CSC) [No.202209370016]. LT was funded by Ghent University (Ghent University Bijzonder Onderzoeksfonds Grant No. 01G01923). The Robson Creek and Litchfield TERN (http://www.tern.org.au) sites are supported by the Australian Government through the National Collaborative Research Infrastructure (NCRIS). TLS fieldwork in Australia was funded by BELSPO (Belgian Science Policy Office) in the frame of the STEREO III programme - project 3D-FOREST (SR/02/355). TLS fieldwork at Wytham Woods was funded through the Metrology for Earth Observation and Climate project (MetEOC-2), grant number ENV55 within the European Metrology Research Programme (EMRP). The EMRP is jointly funded by the EMRP participating countries within EURAMET and the European Union. MD acknowledges support from the UK National Centre for Earth Observation and NERC standard grant NE/P011780/1 “Understanding tree form and function in the tropics”.