GSVA: Generalized Segmentation via Multimodal Large Language Models

MLLMs like GSVA address challenges in GRES by utilizing multiple [SEG] tokens and introducing [REJ] tokens for rejecting empty targets.