In some embodiments, hints are used to make suggestions to the annotator user without making the screen appear cluttered. As used herein, a hint refers to a symbol or marking such as a dot, a line, etc., and is usually more compact than a bounding box and takes up less screen real estate. A hint is placed on or next to an object to indicate that the ML model has determined that there is an object in the vicinity. FIG. 8 is a screenshot illustrating an example user interface for assisted annotation that uses hints. In this example, a dot serves as a hint. Initial bounding boxes of objects that at least meet the confidence level threshold, such as large plane 802, small plane 804, etc. are displayed in full. For predictions that do not meet the confidence level, their bounding boxes are hidden and a dot is placed next to the object as a hint (see, e.g., 806 and 808). When the user clicks on or hovers the cursor over the hint, the corresponding predicted bounding box of the object is revealed, and the user can use the UI tool to adjust or delete the bounding box.

In some embodiments, the hint dynamically changes during the annotation process. FIGS. 9A-9D are screenshots illustrating another example user interface for assisted annotation that uses hints.

FIG. 9A is the original image with all the bounding boxes. The image appears cluttered because there are many overlapping boxes.