Datasets:

Project-AgML
/

WeedsGalore_maize_segmentation

dataset_info:
  features:
    - name: image
      dtype: image
    - name: mask
      dtype: image
    - name: date
      dtype: timestamp[s]
    - name: band
      dtype: string
    - name: split
      dtype: string
  splits:
    - name: train
      num_bytes: 343024917
      num_examples: 780
  download_size: 343074669
  dataset_size: 343024917
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - image-segmentation
size_categories:
  - n<1K

Weedsgalore Maize Segmentation

This dataset provides real-world RGB imagery captured in maize agricultural fields, focusing on weed presence within crop canopies. It includes pixel-level segmentation masks to delineate weeds from maize plants, supporting computer vision research in precision agriculture applications. The dataset contains 780 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

@inproceedings{celikkan2025weedsgalore,
  title={WeedsGalore: A multispectral and multitemporal UAV-based dataset for crop and weed segmentation in agricultural maize fields},
  author={Celikkan, Ekin and Kunzmann, Timo and Yeskaliyev, Yertay and Itzerott, Sibylle and Klein, Nadja and Herold, Martin},
  booktitle={2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  pages={4767--4777},
  year={2025},
  organization={IEEE}
}

https://github.com/GFZ/weedsgalore

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