PUST Cafeteria Food Image Dataset: Real-world Bangladeshi meal-platter images with bounding-box and polygon...

The PUST Cafeteria Food Image Dataset is a real-world image resource developed for object detection and instance segmentation of mixed Bangladeshi cafeteria meals. The source collection comprises 720 meal-platter photographs acquired between 1 August and 5 September 2025 at the Central Cafeteria of Pabna University of Science and Technology, Bangladesh. Images were captured using a Realme GT Master Edition and an OPPO A92 during routine cafeteria operations without standardized illumination, camera position, camera-to-platter distance, background, plate configuration, or food arrangement. Each identifiable food instance was manually annotated in Roboflow using a bounding box and polygon vertices delineating its visible region. Following auto-orientation, resizing, selective augmentation, light class balancing, and quality review, 1,728 images containing 6,297 annotated food instances across 16 classes were retained in the final processed dataset. The dataset was organized into training, validation, and test subsets comprising 1,512, 108, and 108 images, respectively. Annotations are provided in YOLO-compatible TXT and COCO JSON formats, supporting reuse across object-detection and instance-segmentation frameworks. The dataset can be reused for regional food-image analysis, comparative benchmarking, transfer learning, domain adaptation, and intelligent food-service research.