We present SegLLM, a novel multi-round interactive segmentation model that leverages conversational memory of both visual and textual outputs to reason over previously segmented objects and past ...