DeshiFoodBD

Traditional Bangladeshi food image classification has become immensely relevant for a variety of reasons, including restaurant selection, travel destination selection, dietary caloric intake, and cultural awareness. However, this is highly challenging to design an effective and useable traditional labelled (English and Bengali) food dataset of Bangladesh for the research purpose. The ‘DeshiFoodBD’ dataset is presented in this article for traditional Bangladeshi food classification purposes. The food images come from two different sources: 1) web scraping and 2) camera (digital, smartphone). The dataset contains 5425-labelled images of 19 famous Bangladeshi foods such as biriyani, kalavuna, roshgolla, hilsha fish, nehari, and so on. The dataset can be used with a variety of CNN architectures, including ResNet50, YOLO, VGG-16, R-CNN, and DPM.孟加拉传统美食图像分类的研究价值与应用潜力日益凸显，其应用场景覆盖餐厅选型、旅行目的地规划、膳食热量摄入管控以及文化认知推广等多个领域。然而，针对科研场景构建高质量、可复用的孟加拉传统标注（英文与孟加拉语）美食数据集，仍面临诸多挑战。为此，本文提出了面向孟加拉传统美食分类任务的`DeshiFoodBD`数据集。该数据集的美食图像采集自两类渠道：1）网络爬虫抓取；2）数码相机与智能手机实拍。数据集涵盖19种知名孟加拉特色美食，共计5425张标注图像，包含比尔亚尼抓饭（biriyani）、kalavuna、玫瑰奶球（roshgolla）、鲥鱼（hilsha fish）、内哈里咖喱（nehari）等品类。本数据集可适配多种卷积神经网络（Convolutional Neural Network, CNN）架构，例如ResNet50、YOLO、VGG-16、R-CNN及DPM等。