LisaCLIP: Locally Incremental Semantics Adaptation towards Zero-shot Text-driven Image Synthesis | IEEE Conference Publication | IEEE Xplore