← LectureES 667 · Deep Learning

SAME INPUT · DIFFERENT TARGET

Beyond Boxes Lab

Change the question. See how the answer’s structure changes.

ANSWER STRUCTURE

Give a spatial prompt

Or select a stored point

Points select stored visible regions. A drawn box retrieves the stored object box with the greatest overlap. No model runs here.

Choose a phrase Refine the box to a stored mask

These phrases retrieve authored examples. They do not test language understanding.

Illustrative stored examples · Works offline

The course image and overlays are authored teaching material, not measured model predictions. Selected regions are shown; the semantic annotation is incomplete.

Data and provenance

The lecture and lab use the same boxes, polygons, keypoints and depth ordering. Text presets map to stored regions. There is only one dog, so the relational phrase does not demonstrate relational reasoning.

Shared example data (JSON)