1 paper
Xingzhou Pang, Yifan Hou, Junling Wang +1
While Large Vision-Language Models (VLMs) excel at interpolation, they suffer catastrophic failures in systematic generalization, most notably in visual counting. In this work, we…