Evaluating Visual Adapters: MIVPG Performance on Single and Multi-Image Inputs

Details MIVPG experiments across single- and multi-image scenarios. Model uses frozen LLM and Visual Encoder, updating only the MIVPG for efficiency.