Calorie Counter App .
In this project, I trained the YOLOv8 model on 10 classes of food data with 30 images per class. These were all manually labeled via Roboflow and augmented to get a total of >700 images to train with, resulting in a Precision-Recall AUC of ~0.83. The YOLOv8 model was also used to detect LCD screens, which used the bounding boxes to crop the LCD to feed into OCR to extract the weight from the scale. The calories are then calculated with USDA calorie per gram data.