AIDE: Agentically Improve Visual Language Model with Domain Experts

Ming-Chang Chiu, Fuxiao Liu, Karan Sapra +5

The enhancement of Visual Language Models (VLMs) has traditionally relied on knowledge distillation from larger, more capable models. This dependence creates a fundamental bottlene…