The distribution shift due to magnetization differences in magnetometers causes machine learning-based magnetic localization models to be ineffective for the end user. We present a novel post-deployment calibration mechanism to mitigate this issue.