Abstract: Pre-trained Vision-Language Models (VLMs) are often used to tackle the challenging task of Open-vocabulary Segmentation (OVS). To preserve the valuable pre-trained knowledge of VLM-based ...