Ziyu Ma

Towards High-Resolution Visual Perception via Hierarchical Entity Exploration

Ziyu Ma, Shidong Yang, Yuxiang Ji, Yiming Hu, Tongwen Huang, Yong Wang, et al.

High-resolution (HR) image perception remains a key challenge in multimodal large language models (MLLMs), as fine-grained details are often lost when the image is processed as a whole. Existing methods either require training to teach models where to look or heuristically divide…Read more