Understanding Language Prior of LVLMs by Contrasting Chain-of-Embedding

This research introduces VIP and TVI estimators to dissect the language prior in Large Vision-Language Models, offering a diagnostic framework for optimizing...

Level: advanced

By Unknown

Category: research