Datasets:
Geographic source of the images
Hello,
Thank you for releasing this dataset.
I would like to know the geographic origin of the crop images.
Were the images captured in India?
If not, which countries are represented?
Were all images collected through the Farmer.Chat project?
I plan to cite this dataset in my research and would appreciate clarification on its geographic coverage.
Thank you!
Hi Sheshank, thanks for asking, and glad the dataset is useful.
The images are almost entirely from India: 1,052 of the 1,053 distinct images, across 14 states, with Bihar at 789 and Andhra Pradesh at 172. One image is from Ethiopia. The photos were taken between 2 December 2024 and 30 March 2025, India's Rabi season.
Yes, all of them came through Farmer.Chat. Each is a photo a smallholder farmer took on their own plot and sent to the service for advice, which agronomists then reviewed on our annotation platform. Farmer.Chat also runs in Ethiopia, Kenya and Nigeria, but this expert-review batch was drawn almost entirely from India.
We have added country and state columns to annotations.csv and a geographic coverage table to the README, so this is now in the dataset itself. Two things to represent accurately if you cite it: the location is the farmer's registered state, not GPS read from the image (the service records no capture coordinates), and we keep it at state level on purpose, since several states here contain a single image.
Hi, thank you again for the clarification.
I have one more question. Does Farmer.Chat also have an expert-reviewed dataset for crop pest cases (images along with expert annotations or treatment recommendations), similar to this disease dataset? If so, would it be possible to share a subset of that dataset for academic research? Even a small research-only subset would be incredibly helpful.
Thank you for your time!
Hi @15laddoo , thank you for your previous responses.
I'm currently using this dataset for my research, and I was wondering if Digital Green has a larger expert-reviewed crop disease dataset available.
For my experiments, I ideally need around 5,000–10,000 images, with at least 40–50 images per crop–disease class, as the current dataset is too small.
If such a dataset exists, would it be possible to access it for academic research? Even if it isn't publicly available, I'd be happy to use it under any research agreement or licensing terms that you require.
Thank you for your time and consideration!