Food Detection and Recognition with Deep Learning: A Comparative Study

Food detection and recognition involves the use of computer vision and machine learning techniques to identify and classify food items in images or videos. It has numerous applications, such as dietary tracking, nutrition analysis, and inventory management. This research paper presents a comparative study of six deep learning models (SSD (VGG-16), Faster-RCNN (Resnet-50), Faster-RCNN (Mobilenet-V3), Faster-RCNN (Mobilenet-V3_320), RetinaNet (Resnet-50), and YOLOv5) for food detection and recognition. The models' performance is evaluated using three publicly available datasets: School Lunch Dataset, UEC FOOD 100, and UEC FOOD 256. Notably, Faster R-CNN (Mobilenet-V3) achieved mAP of 0.931 in the School Lunch Dataset, while YOLOv5 achieved 0.774 and 0.701 mAP in the UEC FOOD 100 and UEC FOOD 256 Datasets, respectively. YOLOv5 demonstrates comparable results to Faster R-CNN but with a smaller input image size and a larger batch size in food detection.