LeAF Pest Detection Dataset
A large-scale dataset for training and evaluating computer vision models for agricultural pest detection and classification.
Aditya Sengupta, Ana Lucic, Rob Kooper, Vikram Adve
The LeAF Pest Detection Dataset is a comprehensive collection of images and annotations designed to facilitate the development and evaluation of robust computer vision models for identifying and classifying agricultural pests. The dataset features:
- Object Detection and Classification: Images are annotated with bounding boxes around pests, along with corresponding class labels.
- Large Number of Classes: The dataset includes 3,580 distinct agricultural pest classes, representing a wide variety of pests affecting various crops.
- Large Number of Images The dataset contains 35,982 images.
- High-Quality Images: Images were sourced from iNaturalist, a citizen science platform, ensuring a diverse range of real-world conditions, including variations in lighting, background, and pest appearance.
- Data Splits: The dataset is divided into training, validation, and testing sets to support standardized model development and evaluation.
- Annotation Format: Annotations are provided in standard YOLO format. Each image has a corresponding text file containing bounding box coordinates (normalized x_center, y_center, width, height) and class IDs.
Aditya Sengupta, Ana Lucic, Rob Kooper, and Vikram Adve. [LeAF Pest Detection Dataset] [AIFARMS Blog]. 2025. [URL].
This work was supported by the AIFARMS (AI Institute for Future Agricultural Resilience, Management, and Sustainability) National AI Research Institute, funded by the USDA National Institute of Food and Agriculture and the National Science Foundation.