This repository contains all the scripts and resources used for the article *Monitoring plot-level tropical forest canopy structure with automated crown segmentation from low-cost drone imagery *.
It enables full reproducibility of the study, from automatic tree crown segmentation with Detectree2SAM to the statistical analyses and figures presented in the paper.
detectree2SAM/— Python pipeline, Dockerfile and dependencies to run Detectree2SAM and SAM.analysis/— RMarkdown script reproducing the statistical analyses and figures.- Raw data and pre-trained models are hosted externally (see below).
The raw data used in the study (orthophotos, field inventory, segmentation results, and pre-trained models for Detectree2 and SAM) are hosted here:
➡️ Download data & models (DOX ULiège)
Contents of the shared folder:
- Field data (inventory, shapefiles, orthophotos used for validation).
- Pre-trained Detectree2 and SAM models.
- Intermediate segmentation results (to reproduce the R analyses directly if needed).
- GDAL wheel used in requirement.txt
- Docker image (.tar archive)
- Checkpoints containing pre-computed IoU results used in
analysis/Script_analyse_art.Rmd.
These files let you reproduce the statistics and figures without re-running the time-consuming IoU computations. If you prefer to recompute IoU yourself, simply remove the checkpoints and re-run the RMarkdown script.
- Code: MIT License
- Data: Creative Commons Attribution 4.0 International