TeaLeafDiseaseBD: Annotated smartphone images and lesion-level labels of tea leaf diseases from Bangladesh

This dataset contains 13,085 high-resolution images of tea leaves which has been taken using CMF Phone 2 Pro (a 50-megapixel camera phone) in the autumn of the year 2025. The images were collected from the three major tea-producing areas of Sylhet, Sreemangal, and Habiganj, Bangladesh, which ensured there was a wide variety of geography and environments. Every single tea leaf was taken against a black background in order to reduce visual interference and clearly show the disease and the natural leaf variability.

There are seven classes within the data set, consisting of one healthy class and six types of diseases commonly found on tea leaves these are: Red Spider, Brown Blight, Gray Blight, Helopeltis, Algal Leaf Spot, and Red Rust. Disease labeling was done together with professionals from the Agricultural Department, thus, assuring the precision and trustworthiness of the field. The size of all images was set to 1024 × 1024 pixels and they were normalized to that size, which resulted in reproducibility and stable performance across vision models.

Each of the labels of classification besides those that are of diseases at the level of the whole region was also included in the dataset. Lesion regions were marked using an automated computer vision pipeline that had HSV color segmentation, morphological filtering, and connected-component analysis in its composition. This automated method was chosen over manual because it could significantly reduce inter-annotator variability and ensure consistency in lesion segmentation. Annotations contain bounding boxes that are intended to mark biologically relevant regions of interest.

All the created annotations were checked strictly to make sure that background noise and fragmented detections were eliminated. The corresponding annotation files are available in YOLO-compatible plain-text format, which makes it possible to directly use them in object detection workflows. The dataset does not have pre-established