RiceLeafDiseaseBD: A Field-Based Annotated Smartphone Image Dataset of Healthy and Diseased Rice Leaves from Bangladesh

Published: 1 September 2026| Version 3 | DOI: 10.17632/86s4jzj2m4.3
Contributors:
, Toriqul Islam, Smita Saha, Iqbal Hossain Safy, Saiful Bari, Jasmin Akter, Md Momtajur Rahman, Riya Akter, Nahid Hossain Antu, Acramul Haque Kabir, Reefka Fabliha Tulona, Ashraful Islam, M Ashraful Amin

Description

RiceLeafDiseaseBD is a large-scale rice leaf disease dataset containing 9,769 high-resolution images (1024×1024 pixels) with 21,460 bounding box annotations. Images were captured using smartphones (CMF Phone 2 Pro and Google Pixel 5) from rice fields in Gazipur, Dhaka, Bangladesh (≈23.9184° N, 90.4227° E), during the Summer 2025 growing season. Dataset Contents: 6 classes: Healthy, Blast, Brown Spot, Leaf Smut, Rice Tungro, Sheath Blight Original images organized by class folders (EXIF orientations baked in) YOLO format annotation files for object detection (class ID + normalized center, width, height) Visualization images with bounding box overlays, fully aligned with the label files Annotation Protocol (PDF), data collection pipeline and folder structure diagrams Metadata file (image IDs, original and resized sizes, variety, capture device, severity, growth stage). Key Features: Field-based collection with natural backgrounds and lighting Focus on BRRI dhan28 and BRRI dhan29 varieties Manually drawn annotations, expert-validated by plant pathologists and farmers (Cohen's κ = 0.82) Version 3: corrected coordinate extraction (21,460 verified boxes), corrected class IDs, and rotation-consistent images users of earlier versions should migrate Compatible with TensorFlow, PyTorch, YOLOv5/v8 For detailed information, see the README file and Annotation Protocol included in the dataset.

Files

Institutions

Categories

Artificial Intelligence, Machine Learning, Rice, Agricultural Plant, Plant Pathology, Agriculture

Licence