Tree Crown Segment is a QGIS plugin for delineating individual tree crowns in georeferenced RGB orthomosaics. It runs an isolated local inference Engine, writes crown polygons to GeoPackage, and adds the result to QGIS.
Author: Alena D. Nikitina
Affiliation: Isaev Centre for Forest Ecology and Productivity of the Russian Academy of Sciences (CEPF RAS)
- one plugin package for QGIS 3.36+ and QGIS 4;
- English and Russian interface;
- full-raster processing or an area of interest selected from a project, loaded from a polygon file, or drawn on the map;
- automatic projected CRS selection with manual override;
- CUDA acceleration with CPU fallback;
- tile-based processing, cancellation, and resumable checkpoints for large rasters;
- temporary results for parameter exploration or persistent GeoPackage output;
- verified download and installation of the separately packaged Windows Engine and model from inside QGIS.
- Windows 10 or Windows 11, 64-bit;
- QGIS 3.36 or newer, including QGIS 4;
- approximately 13 GB of temporary free space while the Engine is installed;
- an NVIDIA CUDA-capable GPU is recommended; CPU processing is supported but can be substantially slower.
Conda and a separate Python installation are not required.
- Download the Tree Crown Segment plugin ZIP without extracting it.
- In QGIS, open Plugins → Manage and Install Plugins → Install from ZIP.
- Open Tree Crown Segment → System status.
- Click Install/update components and approve the standard Windows administrator prompt after the downloads have been verified.
- Click Check Engine. A successful report contains
"ok": true. - Select a georeferenced RGB raster and run the segmentation.
The plugin downloads only the Engine and model. Source rasters, selected areas, and results remain on the user's computer and are not uploaded. See Installation and first run for detailed steps.
The input must be a readable georeferenced three-band RGB raster with a valid coordinate reference system. The bundled model was evaluated on RGB UAV orthomosaics with a ground sampling distance of approximately 3–8 cm/pixel. Performance outside that range has not been evaluated within this project.
The output is a polygon layer named crowns. Leaving the output path empty
creates a temporary file-backed GeoPackage. Selecting a path creates a
persistent GeoPackage without changing the source raster. Existing output
files are not overwritten.
Parameter definitions, CRS handling, custom model weights, and limitations are documented in Usage and in the plugin's Help tab.
qgis_plugin/treecrownsegment— QGIS interface, validation, map tools, Processing integration, component installation, and Engine process control;engine— raster tiling, model inference, contour conversion, overlap filtering, checkpoints, and GeoPackage writing;- model package — inference-only Mask R-CNN weights distributed separately;
- portable Windows runtime — third-party Python and native dependencies used by the Engine.
The lightweight plugin does not contain the multi-gigabyte runtime or model weights. Each plugin release pins exact component versions, HTTPS download locations, file sizes, and SHA-256 checksums.
If you use Tree Crown Segment, please cite the software. In publications, reports, presentations, and other public outputs, please also cite:
Nikitina, A. D. (2024). Automatic segmentation of tree crowns in pine forests using Mask R-CNN on RGB imagery from UAVs. Forest Science Issues, 7(3), Article 150. https://doi.org/10.31509/2658-607x-202473-150
Machine-readable citation metadata is available in CITATION.cff.
- Installation and first run
- Usage and parameters
- Model card
- Updating and uninstalling
- Engine exit codes
- Changelog
Tree Crown Segment plugin and Engine source code are licensed under the MIT License, Copyright (c) 2026 Alena D. Nikitina.
The Tree Crown Segment model weights are licensed separately under CC BY 4.0. The portable Engine contains third-party components that retain their own licenses; see Third-party components.