Mangifera-30: A Balanced Image Dataset of 30 Mango Varieties for Computer Vision

Mangifera-30 is a comprehensive, balanced image dataset comprising 30 distinct mango (Mangifera indica L.) varieties and one “Others” class containing non-mango fruit and vegetable images. The dataset was created by integrating four publicly available, open-licensed sources, with full attribution and adherence to reproducibility and ethical data-sharing principles.

Each class contains 500 training and 100 testing RGB images, totaling 18,600 high-quality samples. Images are organized into train and test directories, each with class-specific subfolders, and standardized in resolution and format for seamless use in computer vision pipelines.

The 30 mango varieties are: Amrapali, Anwar RatooI, Ashshina Classic, Ashshina Zhinuk, Banana, Bari-4, Bari-7, Bari-11, Black Chaunsa, Dosehri, Fazlee, Fazlee Shurmai, Gobindovog, Gopalvog, Gouromoti, Harivanga, Himshagor, Kanchon Langra, Katimon, Khirshapat, Langra, Mollika, Nilambori, Ranivog, Rupali, Shada, Sindhri, Summer Bahisht, Sundari, and White Chaunsa.

The “Others” class contains various non-mango images to enhance model generalization and robustness.

Mangifera-30 is suitable as a benchmark for fruit classification, variety identification, agricultural AI, phenotyping, deep learning evaluation, and reproducible computer vision research in the agricultural domain.