Food Recognition and Segmentation Using Detectron2 Framework

Food recognition is the process of identifying food categories present in the input image. Segmentation focuses on outlining the boundaries of each identified food category. This is especially useful for nutrition tracking, which aids people in analyzing and improving their diets. Detectron2 is an open-source library developed by Meta that offers an extensive range of cutting-edge algorithms for object detection and instance segmentation tasks. In this research paper, we utilize Detectron2 to adapt and train a model for food recognition and segmentation based on the Mask-RCNN architecture. The results of the trained model show that it is viable to use Detectron2 for this particular task, however, there is potential for additional work regarding the evaluation and optimization of the model.