VGA-BenchV2 Scales Video Aesthetic Annotations by 13x
August 27, 2026
VGA-BenchV2 expands video evaluation via 36,000 new task-level annotations across 52 sub-dimensions. It utilizes a hybrid evaluator architecture, including VAQA-Net for continuous aesthetic scoring, to assess both generation quality and aesthetic value.
HOW THIS AFFECTS YOU
●
researcherYou can leverage this significantly larger, human-aligned dataset to train more precise video quality evaluators.
●
designerThis benchmark helps define higher aesthetic standards for generative video models.