As shown in the resulting image in FIG. 5B, boxes 502 and 504 are adjusted to cover the entire objects, a new bounding box 508 is added around vehicle 506, and object 510 has been reclassified as a vehicle (shown with a red box). Since the user does not need to draw each box from scratch, this annotation technique improves the speed at which an annotator can process an image.

FIGS. 6A-6B are screenshots illustrating another example user interface for assisted annotation. In this example, the initial predictions are shown in FIG. 6A. The annotator user is not required to click on a box in order to edit the box. Rather, the UI tool allows the user to activate a bounding box (e.g., making the box editable) by using the cursor. In this case, when the user moves the mouse such that the cursor is over the object or the bounding box (e.g., when the cursor hovers over the object or the bounding box), the bounding box is activated and ready to be edited. The adjusted result is shown in FIG. 6B, where the user has adjusted the bounding box sizes for planes 602 and 604, as well as deleting box 606 and resizing box 608. Once the user has completed the annotation, he can submit the result by clicking on a submit button (not shown).