A Comprehensive Image Dataset of Vegetables Grown in Bangladesh
Description
This dataset comprises a total of 4,730 high-quality JPG images, including 1,877 raw, unprocessed images, representing 42 distinct vegetable categories commonly found in Bangladesh. Each vegetable category is uniquely captured to ensure a well-balanced and diverse collection, making the dataset suitable for machine learning applications such as image classification, object detection, and agricultural analysis. The vegetable classes include: Arum Lobe, Ash Gourd, Beetroot, Bitter Melon, Bottle Gourd, Broccoli, Cabbage, Capsicum, Carrot, Cauliflower, Chives Onion, Chili, Coconut, Coriander, Cucumber, Eggplant, Elephant Foot Yam, Flat Bean, Garlic, Ginger, Gooseberry, Green Papaya, Green Spinach, Jicama, Kohlrabi, Lime, Malabar Spinach Seed, Okra, Onion, Plantain, Pointed Gourd, Potato, Pumpkin, Radish, Radish Leaves, Red Amaranth, Shaluk, Snake Gourd, Taro, Tomato, Yardlong Bean, and Zucchini. This curated dataset aims to support researchers and developers in the fields of computer vision, agriculture, and food technology by providing a reliable and diverse set of labeled images for model training, testing, and validation.
Files
Steps to reproduce
To compile this dataset, vegetable samples were sourced from various street stalls and local markets in the Mirpur area of Dhaka City, Bangladesh. The data collection process focused on capturing diverse and commonly available vegetable types in realistic settings. All images were taken using a Poco F3 Android smartphone, configured to its highest resolution settings (up to 48 MP), ensuring detailed and high-quality captures. Photographs were taken under natural daylight conditions without any artificial lighting or flash, in order to preserve the authentic visual characteristics of each vegetable, including color, texture, and shape. Each sample was carefully positioned and photographed from different angles to capture unique features and reduce visual redundancy. A total of 4,730 JPG images were collected, of which 1,877 are raw, unprocessed images. After collection, all images were manually reviewed to ensure clarity and relevance, then categorized into 42 distinct vegetable classes. The dataset was organized into class-specific folders, and no preprocessing or augmentation was applied to the raw images to maintain original data integrity. This method offers a straightforward and reproducible approach for researchers aiming to create similar image datasets under real-world conditions using accessible tools and environments.