UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models | Read Paper on Bytez