1) Are suited to the goal of research - i.e bounding boxes to identify multiple classes at once, object detection to see only if a species is present

2) Are uniform - do not mix annotation types

3) Consistency in annotation targets - i.e all fish present in two or more similar/identical frames are annotated

4) All subjects of interest are annotated - i.e all fish in the frame are found and annotated

5) Clear labels that correctly identify subject of interest

6) Annotations have been peer reviewed

7) Annotations are saved in a file type compatible with the desired machine learning model.