Project-AgML/image_safari_plant_part_segmentation
Image Safari Plant Part Segmentation Plant-part segmentation of field photographs from the Image Safari project, a large crop image collection from sites across sub-Saharan Africa. In 1,361 smartphone images of 7 crops, QED annotators outlined individual leaves, stems (including banana pseudostems and suckers), flowers (buds, open flowers, inflorescences, panicles, spikes), fruits (pods, grains, tubers, capsules, banana bunches) and broadleaf/narrowleaf weeds in Roboflow. The… See the full description on the dataset page: https://huggingface.co/datasets/Project-AgML/image_safari_plant_part_segmentation.
Image Safari Plant Part Segmentation
Plant-part segmentation of field photographs from the Image Safari project, a large crop image collection from sites across sub-Saharan Africa. In 1,361 smartphone images of 7 crops, QED annotators outlined individual leaves, stems (including banana pseudostems and suckers), flowers (buds, open flowers, inflorescences, panicles, spikes), fruits (pods, grains, tubers, capsules, banana bunches) and broadleaf/narrowleaf weeds in Roboflow. The annotation_type column separates fully manual plant-part annotations (761 images) from model-assisted pseudo-instance annotations (600 images); some images were annotated in both, and appear once per annotation batch (annotation_set). Each row has the image, a semantic mask (0 = background, then the class ids; larger objects are painted first so small parts stay visible), per-object bounding boxes and classes, the crop, and the original per-image COCO annotation file verbatim in coco_json (instance polygons, original category names). Images that the source stored with an EXIF rotation are written upright so they line up with the annotations; all others keep their original bytes. The dataset contains 1,361 images with 240,726 bounding box annotations across 17 categories.
This dataset is indexed on https://project-agml.github.io/ as part of the AgML python library.
Citation
@article{Mutuvi_2026,
title = {Artemis and Image Safari - crop imagery datasets spanning globally critical crop species},
author = {Mutuvi, Stephen and Guerena, David and Zych, Marcin and Henday, Shekinah and Girma, Ellena and Mungubariki, Tumaini and Agesa, Beverly and Zochowski, Marek and Ciolek, Daniel and Lazowik, Michael and Chen, Jiehua and Goeke, Leah and Omwandho, Justine and del Palma, Grace and Malabi, Jackson and Phomebeya, Sithembinkosi and Sanena, Maria and Marcos, Juliana T. C. and Ghandi, Harish and Nas, Mark and Rathore, Abhishek and Mendes, Thiago and Yadav, Shailesh and Adjah, Kossi Lorimpo and Woltering, Lennart and Lekasio, Elice and Mushi, Bryton and Abraham, Leah and Katunzi, Gaudensia and Remy, Sekou L. and Siyavora, Tavonga and Selvaraj, Michael and Casas, Jorge and Boukar, Ousmane and Ongom, Patrick and Nakato, Gloria Valentine and Mwanje, Gerald and Agre, Paterne and Laporte, Marie-Angélique and Asefa, Teshale and Odama, Roy and Mukankusi, Clare and Wu, William},
journal = {agriRxiv},
publisher = {CABI Publishing},
year = {2026},
month = sep,
doi = {10.31220/agriRxiv.2026.00491},
url = {https://doi.org/10.31220/agriRxiv.2026.00491}
}This dataset was reformatted from its original format to match HuggingFace standards.
