When there are lots of objects in an image, displaying all the initial predicted bounding boxes or all the qualifying predicted bounding boxes using the technique shown above may result in a cluttered image, which can cause eye fatigue for the annotator user and reduce accuracy and throughput. Thus, in some embodiments, rather than displaying all the bounding boxes, only a portion of the bounding boxes are selectively displayed and the rest of the bounding boxes are hidden.

FIG. 7 is a screenshot illustrating another example user interface for assisted annotation. In this example, N qualifying predicted bounding boxes that are nearest to the cursor (N being a predefined natural number and is set to 5 in this case) are activated (e.g., displayed and made editable). In some embodiments, an area around the cursor location (e.g., a circle of a certain diameter or a square of a certain size centered at the cursor location) is determined and all objects within the area are activated and made editable.