DigiGreen/Crop_Disease_Images

Crop Disease Expert Annotations

1,026 expert annotations over 989 crop photographs, covering pests, diseases and nutrient deficiencies across 74 crop types. The images are included in this repository.

Every image is a photo taken by a smallholder farmer on their own plot and sent to Farmer.Chat, an AI advisory service run by Digital Green. Agronomists then reviewed each photo on Digital Green's annotation platform. Nothing here is scraped, staged, or lab-photographed.

Contents

• images/ - 989 JPEG files, named 00001.jpg upward
• annotations.parquet - the annotations, one row per expert review
• annotations.csv - the same table as CSV

Column	Description
image_file	Path to the image within this repository
crop	Crop or plant identified by the expert
diagnosis	Pest, disease, or nutrient deficiency name(s); Healthy if none
details	Expert's observations on visible symptoms
country	Country the farmer who sent the photo is registered in
state	State or province the farmer is registered in

Multiple diagnoses per image are semicolon-separated.

Usage

from datasets import load_dataset
from huggingface_hub import snapshot_download

ds = load_dataset("DigiGreen/Crop_Disease_Images", split="train")
root = snapshot_download("DigiGreen/Crop_Disease_Images", repo_type="dataset")

row = ds[0]
print(row["crop"], row["diagnosis"])
image_path = f'{root}/{row["image_file"]}'

Stats

	Count
Total annotations	1,026
Distinct images	989
Diseased	723
Healthy	303
Unique crops	74

37 images were reviewed by more than one expert, which is why the row count exceeds the image count.

Geographic Coverage

Photos were taken between 2 December 2024 and 30 March 2025, which is India's Rabi (winter) season. Counts below are distinct images.

Country	State	Images
India	Bihar	738
India	Andhra Pradesh	161
India	Telangana	36
India	Jharkhand	21
India	Odisha	17
India	Uttar Pradesh	5
India	Karnataka	3
Ethiopia	Addis Ababa	1
India	Gujarat	1
India	Haryana	1
India	Kerala	1
India	Madhya Pradesh	1
India	Maharashtra	1
India	Rajasthan	1
India	West Bengal	1

India accounts for 988 of 989 images (99.9%) across 14 states, with Bihar alone at 75%. Farmer.Chat also operates in Ethiopia, Kenya and Nigeria, but this expert-review batch was drawn almost entirely from India.

country and state come from the registered location of the farmer who sent the photo, so they say where that person farms. Finer administrative levels are withheld, because several states here contain a single image and a district would identify an individual farm.

Privacy

Every image was reviewed for personal information before publication, by two vision models independently plus a human pass over every case where the two disagreed. Images were removed where they contained a person's face, a name or phone number, a chat screenshot, or a location stamped into the picture by the camera app the farmer used.

64 images were removed on those grounds. Some camera apps overlay the capture coordinates, village name and date onto the photo, which is far more precise than the state-level location published here, so those images were dropped rather than cropped. Screening reduces this risk but does not eliminate it: a faint or partly legible stamp may remain. If you find one, please open a discussion and it will be removed.

Earlier versions of this dataset delivered images as links to a cloud storage bucket, and the filenames in those links carried an internal message identifier. Both are gone. Images are now served from this repository under sequential filenames that carry no information about the farmer, the conversation, or the original record.

The images carry no EXIF metadata. Farmer and reviewer identities are not included, and no free-text farmer message is published.

Changelog

July 2026. Images moved into the repository as images/00001.jpg upward and the image_url column was replaced by image_file. Code that read image_url needs updating. Ran a full privacy screen over every image and removed 64 that carried personal information; four rare diagnosis labels had only one image each and went with them.

Earlier in July 2026. Added country and state.

Top 10 Crops

Crop	Count
Chilli	124
Potato	102
Brinjal	87
Maize	77
Mango	65
Tomato	48
Papaya	44
Wheat	39
Mustard	34
Bean	32

Top 10 Diagnoses

Diagnosis	Count
Healthy	303
Leaf Curl Virus	82
Thrips	67
Aphids	58
Fall Armyworm	52
Powdery Mildew	48
Mites	47
Shoot and Fruit Borer	33
Anthracnose	32
Mosaic Virus	25

Source

Annotations were collected by Digital Green as part of the Farmer.Chat project, an AI-powered agricultural advisory platform serving smallholder farmers.

License

This dataset is released under CC-BY-4.0. You are free to share and adapt the data for any purpose, provided you give appropriate credit to Digital Green.

Citation

@dataset{digigreen_crop_disease_2026,
  title={Crop Disease Expert Annotations},
  author={Digital Green},
  year={2026},
  url={https://huggingface.co/datasets/DigiGreen/Crop_Disease_Images},
  license={CC-BY-4.0}
}