Quickstart
Run tree detection on a single orthomosaic in a few steps.
Prefer a hosted environment?
Our Google Colab tutorial runs the full detect + segment pipeline on a sample orthomosaic on a free T4 GPU and visualizes the results — no local install needed. Its setup is adapted for Colab, and you can reuse its cells in your own notebook to run inference on your own data. Colab is slower and can't handle large orthomosaics, so it's best for testing or small datasets; for production use, install CanopyRS locally.
Sample raster
A small test raster is included in the repository at assets/20240130_zf2tower_m3m_rgb_test_crop.tif. You can use it to try the commands below without needing your own data.
Using a preset configuration
CanopyRS ships with preset pipelines. The fastest way to get started is to use one of them directly via infer.py.
Single raster/orthomosaic input (-i):
python infer.py -c <CONFIG_NAME> -i <PATH_TO_TIF> -o <PATH_TO_OUTPUT_FOLDER>
Folder of already tiled geo-referenced images (-t):
python infer.py -c <CONFIG_NAME> -t <PATH_TO_TILES_FOLDER> -o <PATH_TO_OUTPUT_FOLDER>
Command-line arguments
| Argument | Description |
|---|---|
-c |
Config name (folder name under canopyrs/config/, see Presets for a list of predefined configs.) |
-i |
Input path to a single raster/orthomosaic |
-t |
Input path to a folder of geo-referenced .tif tiles |
-o |
Output path |
Understanding the output
The output folder will contain one subfolder per component that ran, containing output files such as:
- GeoPackage (
.gpkg) — for example predicted tree polygons with scores - COCO JSON — predictions in COCO format (used internally between components, can also be used to visualize per-tile predictions, see TODO)
If the chosen pipeline configuration produced a GeoDataFrame containing polygon results, it will be present at the root of your output folder.
Choosing the right preset
See Presets for full details.