I have observed a possible bug when downloading datasets in the coco format from Roboflow. When creating a new object detection project, the user needs to specify the name of at least one class (eg cars). If this class is later deleted using the “modify classes” feature, it is still present in the coco dataset json categories, resulting in the wrong number of classes.
This tag is also applied as a supercategory for later classes added in labelling.
For example, I have two classes in this project after deleting the initial class “car-plate”:
However, the json file has the following 3 classes under categories:
categories":[{"id":0,"name":"car-plate","supercategory":"none"},{"id":1,"name":"bus","supercategory":"car-plate"},{"id":2,"name":"vehicle","supercategory":"car-plate"}]
Is this behavior intended?