Browse Datasets

KaraAgro AI Maize

Annotated Maize Disease Images from Ghana

object-detection image-classification computer-vision 16,552 images (~1.5GB) CC0 1.0
Access Dataset

Description

Over 16,000 maize crop images with bounding-box annotations for diseases including maize streak and armyworm damage, curated by KaraAgro AI in Ghana. Released CC0; canonical data lives on Harvard Dataverse (doi:10.7910/DVN/CXUMDS).

Quick Start

Python 
# Download from Harvard Dataverse (CC0)
#   https://doi.org/10.7910/DVN/CXUMDS
# then load the images + bounding-box annotations with your CV framework
from torchvision.io import read_image

Project Ideas

Disease Detector

Train a YOLO or transformer-based detector to localise armyworm damage and frass on maize leaves.

Intermediate

Imbalance Study

Evaluate the detector on a heavily imbalanced disease subset and practise targeted data augmentation.

Intermediate

Ethical Considerations

  • Models must not be presented to farmers as definitive guarantees without agronomic oversight.
  • Camera-quality differences across regions can teach artefact bias rather than biology.