[2409.16278] Adapting Vision-Language Model with Fine-grained Semantics for Open-Vocabulary Segmentation