Computer vision

Methods that extract information from images, video or other visual measurements.

Tasks include detection, segmentation, tracking and interpreting actions. A visual prediction still needs an evaluation rule: detecting a cup is different from deciding whether a manipulation succeeded. The evidence required depends on the question, and a single frame may be insufficient for an event that unfolds over time.

In practice

Checking whether an object stays supported after release requires viewing the relevant sequence, not just locating it in one image.