A Training-Free, Task-Agnostic Framework for Enhancing MLLM Performance on High-Resolution Images

Jaeseong Lee, Yeeun Choi, Heechan Choi +2

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in vision-language understanding, reasoning, and generation. However, they struggle with tasks re…