We propose a VLM-assisted dual-space adaptation framework to improve unsupervised domain adaptive hashing by calibrating pseudo-labels and decoupling feature and Hamming space alignment.