FoodBD: A Polygon-Annotated Meal Image Dataset of Bangladeshi Cuisine with Visual and Nutritional Labels

FoodBD dataset, a novel food image dataset comprising 3,523 smartphone-captured meal images of Bangladeshi cuisine. Each image is annotated with polygon-based segmentation of individual food items spanning 67 categories. Additionally, 1,837 images include detailed nutritional information: carbohydrate, protein, fat, fiber, calorie, and glycemic load, estimated by a nutrition expert. FoodBD enables a wide range of tasks including multi-label classification, object detection, semantic segmentation, and nutrition estimation.

FoodBD 数据集（FoodBD dataset）是一款新颖的餐饮图像数据集，包含3523张由智能手机采集的孟加拉料理餐食图像。每张图像均针对67个类别下的单个食材完成基于多边形的分割标注。此外，其中1837张图像还附带由营养专家估算得到的详细营养信息，涵盖碳水化合物、蛋白质、脂肪、膳食纤维、热量及血糖负荷。FoodBD 可支持多标签分类、目标检测、语义分割以及营养估算等多种任务。