About

The CWD30 dataset contains high-resolution images of 20 weed and 10 crop species, capturing diverse growth stages, viewing angles, and environmental conditions from various agricultural fields and in lab grown plants. For download, refer to the download page.

  • CWD30
    • The CWD30 dataset comprises a collection of 219,770 images that encompass 10 crop classes and 20 weed classes. These images capture various growth stages, multiple viewing angles, and diverse environmental conditions. The detailed hierarchical taxonomy information can be found here.
    • Sample Images
    Magnifiable Image

  • CWD30-S (Coming Soon!)
    • CWD30-S is an extension of the CWD30 dataset, designed for semantic segmentation tasks. Following annotations will be provided:
      Sample Images
      Magnifiable Image

      • Semantics:
        • Pixel-wise semantic masks, where label ids correspond to:
          • background (0)
          • crop (1)
          • weed (2)
        • The crops will have further sub-annotations available:
          • legumes (3)
          • grains (4)
          • oilseeds (5)
        • For weeds, we'll have further sub-annotations available:
          • grasses (6)
          • sedges (7)
          • broad-leaves (8)
          • unknown (9)
      • Plant Instances: Pixel-wise instance masks for crops and weeds, where ids > 0 correspond to distinct instances.

Announcement

2024/06/06 The paper is currently under review.
2024/05/11 The CWD30-S data will be released on link.
2023/05/17 The paper link (v1): arXiv.

Citation

@article{ilyas2025cwd30,
  title={CWD30: A new benchmark dataset for crop weed recognition in precision agriculture},
  author={Ilyas, Talha and Arsa, Dewa Made Sri and Ahmad, Khubaib and Lee, Jonghoon and Won, Okjae and Lee, Hyeonsu and Kim, Hyongsuk and Park, Dong Sun},
  journal={Computers and Electronics in Agriculture},
  volume={229},
  pages={109737},
  year={2025},
  publisher={Elsevier}
}