The dataset annotations (json file) has 90 classes ids (+ background_class), but only 80 classes actually. It can lead to confusion regards