Training-Free Test-Time Adaptation via Shape and Style Guidance for Vision-Language Models | Read Paper on Bytez