GSVA: Generalized Segmentation via Multimodal Large Language Models