scvance/TomatoWUR

About

Official implementation of TomatoWUR dataset:

An annotated dataset of tomato plants to quantitatively evaluate segmentation, skeletonisation, and plant trait extraction algorithms for 3D plant phenotyping

The dataset is related to the paper: 3D plant segmentation: Comparing a 2D-to-3D segmentation method with state-of-the-art 3D segmentation algorithms

Installation

This software is tested on Python 3.11. To install the dependencies, run:

pip install -r requirements.txt

Usage

Make sure to extract and download the dataset, this will be done automatically if path can not be found:

python3 wurTomato.py --visualise 0

For more examples have a look at the example_notebook.ipynb

Settings are described in config file

Training Bundles In 2D-to-3D_segmentation

This dataset repo is also used as a submodule inside the parent 2D-to-3D_segmentation project for Pointcept experiments.

That project generates non-destructive training bundles under:

data/TomatoWUR/ann_versions/<version-name>/

For the original frame-wise partial data, the bundle contains:

json/train.json
json/val.json
json/test.json

For the newer trajectory experiments, the same builder also writes:

json/train_trajectories.json
json/val_trajectories.json
json/test_trajectories.json

Those trajectory manifests preserve frame order inside each trajectory so the loader can expose sequence boundaries to a future recurrent model.

Example command from the parent repo with Docker Compose:

docker compose -f ../docker-compose.yaml run --rm \
  -v /path/to/TomatoWUR_trajectory:/data/TomatoWUR_trajectory \
  interactive python3 /workspace/plant3d/TomatoWUR/data/TomatoWUR/build_partial_ann_version.py \
  --annotations-root /data/TomatoWUR_trajectory/annotations_trajectory_sensor \
  --point-clouds-root /data/TomatoWUR_trajectory/point_clouds_trajectory_sensor \
  --version-name trajectory-sensor-plant \
  --pairing-mode strict \
  --split-unit plant \
  --sequence-delimiter _sensor_ \
  --train-ratio 0.8 \
  --val-ratio 0.1 \
  --test-ratio 0.1 \
  --seed 123 \
  --materialize-mode copy \
  --dry-run

Use --split-unit plant for sequential experiments so all trajectories from the same plant stay in the same split. Detailed path conventions and usage are documented in data/README.txt in this repo and in the parent project readme.md.

Citation

@article{VANMARREWIJK2025111852,
title = {TomatoWUR: An annotated dataset of tomato plants to quantitatively evaluate segmentation, skeletonisation, and plant-trait extraction algorithms for 3D plant phenotyping},
journal = {Data in Brief},
volume = {61},
pages = {111852},
year = {2025},
issn = {2352-3409},
doi = {https://doi.org/10.1016/j.dib.2025.111852},
url = {https://www.sciencedirect.com/science/article/pii/S2352340925005773},
author = {Bart M. {van Marrewijk} and Tim {van Daalen} and Katarína Smoleňová and Bolai Xin and Gerrit Polder and Gert Kootstra},
}

Related research

2Dto3D segmentation paper

Funding

This research is part of AgrifoodTEF: Test and Experiment Facilities for the Agri-Food Domain (101100622)

README