1 briefs tagged #visual-question-answering.
The framework highlights image regions and text evidence before answer generation.