A vision model only sees what its training data taught it. Identify the right object from a tiny pixel render — before the scan completes!