ashu010/agridrone-data

AgriDrone — Crop Disease Detection Dataset

Full dataset collection used for training the AgriDrone crop-disease detection system (21-class YOLOv8n-cls classifier, 15 wheat + 6 rice diseases). 75,010 images, 24 folders, 23.6 GB.

Code repo: https://github.com/Ashut0sh-mishra/agri-drone

Folder layout

Training splits (5 folders — used to train the model)

Folder	Files	Purpose
train/	10,144	Main training split
val/	1,655	Validation split
test/	1,655	Test split (used for the Config A/B/C ablation)
train_orig/	4,987	Original pre-augmentation training set
val_orig/	1,247	Original pre-augmentation validation set

Wheat disease classes (11 folders)

Folder	Files
wheat_aphid/	300
wheat_blast/	300
wheat_healthy/	600
wheat_leaf_rust/	300
wheat_smut/	300
wheat_yellow_rust/	300
fusarium_head_blight/	300
leaf_blight/	300
powdery_mildew/	300
septoria/	300
tan_spot/	300

Rice disease datasets (3 folders)

Folder	Files	Source
Rice_Leaf_AUG/	3,829	Augmented rice leaf set
rice-diseases-v2/	10,346	Roboflow rice disease v2
rice-diseases-zoa8l/	2,589	Roboflow rice disease (zoa8l)

External benchmarks (3 folders)

Folder	Files	Description
PDT dataset/	19,049	Plant Disease Treatment dataset
plantdoc/	196	Original PlantDoc benchmark
plantdoc-v3/	1,552	PlantDoc v3

Raw detection set (1 folder)

Folder	Files	Description
data/	14,154	Roboflow-sourced wheat detection dataset (bounding boxes)

Usage

Python — download a subset

from huggingface_hub import snapshot_download

# Training splits only (enough to reproduce the 21-class model)
path = snapshot_download(
    repo_id="ashu010/agridrone-data",
    repo_type="dataset",
    allow_patterns=["train/**", "val/**", "test/**"],
)

CLI — use the fetch script

From the GitHub repo:

python scripts/fetch_data.py --preset training   # train/val/test splits
python scripts/fetch_data.py --preset wheat      # all 11 wheat classes
python scripts/fetch_data.py --preset rice       # 3 rice datasets
python scripts/fetch_data.py --preset external   # PlantDoc + PDT
python scripts/fetch_data.py                     # everything (25 GB)

Direct image URL (for dashboards / web apps)

Every image has a public resolve URL — no auth needed:

https://huggingface.co/datasets/ashu010/agridrone-data/resolve/main/<folder>/<filename>

Example:

<img src="https://huggingface.co/datasets/ashu010/agridrone-data/resolve/main/wheat_aphid/wheat_aphid_0001.jpg">

Use these URLs directly in the AgriDrone frontend dashboard for live data display.

License

MIT for the curation layer. Individual source datasets retain their original licenses — see each folder for upstream attribution.

Citation

See the AgriDrone repo: https://github.com/Ashut0sh-mishra/agri-drone