-
-
Notifications
You must be signed in to change notification settings - Fork 57
Splitting Data
Testing
Converting the format of the annotations is sometimes not enough for the annotations to work with the desired deep learning framework, such as YOLO. In YOLO, for example, the class ids must start from zero and be continuous. For example you have 3 classes: Squirrel, Nut, Bird then you must have 3 class ids: 0,1,2. The number of classes is 3 and the order of the class names should be specified in the YAML file as [Squirrel, Nut, Bird]. If you used class id 3 for Bird then you would get an error when you train the model.
Other models such as Detectron require the class ids to start at 1.
To help with getting datasets in a model ready state, PyLabel has reindexing features to ensure that class ids are continuous and start from the desired index.