Ruiwen Xu

We propose a novel hybrid of position embedding to improve the length generalization ability of Vision-Language Models.