---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: mask
    dtype: image
  - name: image_series
    dtype: string
  - name: split
    dtype: string
  splits:
  - name: train
    num_bytes: 888005233
    num_examples: 92
  download_size: 888031817
  dataset_size: 888005233
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: cc-by-4.0
task_categories:
- image-segmentation
size_categories:
- n<1K
---
# Alpine Grassland Root Segmentation

92 root images (2550x2196) with manual binary root/background masks from the 'Grassland' dataset, collected in alpine grasslands during a study on root growth and soil function. It is one of the nine root imaging datasets in a benchmark of 21 Transformer and ConvNet architectures for root segmentation. Images keep their image series (BidmerN, BidmerS, Phytotron2021) and the record's U-Net training/validation split. The dataset contains 92 images with pixel-level mask annotations.

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

The original train/test/val split has been preserved in the `split` column.

## Citation

```bibtex
@article{smith2026systematic,
  title={A systematic comparison of transformers and ConvNets for root segmentation across nine datasets},
  author={Smith, Abraham George and Lamprinidis, Sotiris and Seethepalli, Anand and York, Larry M. and Han, Eusun and Möhl, Patrick and Boulata, Kyriaki and Thorup-Kristensen, Kristian and Petersen, Jens},
  journal={Plant Methods},
  volume={22},
  pages={52},
  year={2026},
  publisher={BioMed Central}
}
```


The dataset itself can be cited as:

Möhl, P., von Büren, R. S., &amp; Hiltbrunner, E. (2025). *Growth of alpine grassland will start and stop earlier under climate warming (dataset)* [Dataset]. figshare. https://doi.org/10.6084/M9.FIGSHARE.20440497.V2


*This dataset was reformatted from its original format to match HuggingFace standards.*