Project-AgML/pheno4d_point_cloud_segmentation

Pheno4D Point Cloud Segmentation

A dataset for point cloud semantic segmentation of maize and tomato plants. The dataset contains 126 labeled 3D scans across two species, tomato (single label scheme) and maize (dual label scheme: collar-based and tip-based organ boundaries), with per-point labels identifying plant organs (e.g. stem, leaf).

Each sample includes:

• points: an (N, 3) array of x, y, z point coordinates, N varies per scan
• mask: an (N, 1) array of per-point organ labels (primary label scheme)
• mask_2: an (N, 1) array of per-point organ labels (secondary label scheme, maize only; -1 where not applicable)

This dataset is indexed on https://project-agml.github.io/ as part of the AgML python library.

Note: this is AgML's first point cloud dataset. Naming follows the mask convention used for 2D semantic segmentation datasets in this organization, adapted for 3D point data (stored as Array2D with a dynamic first dimension rather than image masks).

Citation

@article{schunck2021pheno4d,
  title={Pheno4D: A spatio-temporal dataset of maize and tomato plant point clouds for phenotyping and advanced plant analysis},
  author={Schunck, David and Magistri, Federico and Rosu, Radu Alexandru and Corneli{\ss}en, Anne and Chebrolu, Nived and Paulus, Stefan and L{\'e}on, Jens and Behnke, Sven and Stachniss, Cyrill and Kuhlmann, Heiner and Klingbeil, Lasse},
  journal={PLOS ONE},
  volume={16},
  number={8},
  pages={1--18},
  year={2021},
  publisher={Public Library of Science},
  doi={10.1371/journal.pone.0256340}
}

Original dataset: https://www.ipb.uni-bonn.de/data/pheno4d/

This dataset was reformatted to match HuggingFace standards as part of AgML's point cloud dataset support.