Towards Seamless Adaptation of Pre-trained Models for Visual Place Recognition