Too Long; Didn't Read

This section defines the task of zero-shot visual question answering (VQA) and explores the use of pre-trained vision-language models (PVLMs) like BLIP-2, highlighting its Querying Transformer component for bridging the modality gap in cross-modal understanding.