Project-AgML/OpenEar_projection_segmentation

Openear Projection Segmentation

This dataset provides real-world RGB images of maize ears in a field environment at Hongqi Base, Hainan, China, captured using a ground-based Raspberry Pi HQ camera system. Collected over a period from March 2025 to January 2026, it offers a longitudinal resource for semantic segmentation tasks in agricultural phenotyping under natural field conditions. The dataset contains Augmented and Raw versions. The Augmented split contains 1,048 images with pixel-level mask annotations.

The Raw split contains 248 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

@article{fan2026openear,
  title={OpenEar: an ultra-affordable, high-throughput, and accurate maize ear phenotyping system},
  author={Fan, Shaoqi and Li, Guoji and Bahitwa, Revocatus and Jia, Zhiguo and Zhang, Hongwei and Shao, Jinghong and Yu, Qiuying and Chen, Xiaoran and Qian, Yiheng and Xu, Mingchi and Zhu, Linlin and Wang, Hai},
  journal={Plant Methods},
  volume={22},
  pages={26},
  year={2026},
  publisher={BioMed Central}
}

Fan, S. (2025). Datasets for OpenEar model training (Version 2). figshare. https://doi.org/10.6084/m9.figshare.26282563.v2

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