Search papers, labs, and topics across Lattice.
This paper introduces Structured All-Mask Prediction (STAMPlus), a novel approach to MLLM-based segmentation that effectively addresses the segmentation trilemma of performance, dialogue preservation, and inference speed. By decoupling autoregressive dialogue from non-autoregressive mask prediction and utilizing a unified target list with explicit IDs, STAMPlus achieves state-of-the-art segmentation performance while significantly reducing latency from 13.50s to 5.16s. The method also enhances multimodal instruction following and enables open-vocabulary, instance-aware segmentation, demonstrating its versatility across various applications.
STAMPlus not only slashes segmentation latency but also excels in open-vocabulary and instance-aware tasks, redefining the capabilities of MLLM-based segmentation.
MLLM-based segmentation faces a core segmentation trilemma: high segmentation performance, preserved dialogue ability, and fast inference. Embedding-prediction methods may disrupt language modeling through pixel-level objectives, whereas next-token generation is inefficient for dense masks. We propose All-Mask Prediction, decoupling autoregressive dialogue from non-autoregressive mask prediction. Its binary instantiation, STAMP (Simultaneous Textual All-Mask Prediction), emits an in-vocabularytrigger, fuses image-aligned mask tokens with corresponding patch features, and uses hybrid attention to classify all tokens as foreground or background in one pass. It thereby combines strong referring and reasoning segmentation with preserved multimodal ability and efficient inference. However, binary masks cannot retain multiple semantic or instance identities without repeated target-specific predictions. We therefore propose Structured All-Mask Prediction and develop STAMPlus. It generates a target list with explicit IDs and optional boxes, binds these IDs to a shared multi-class mask space, and jointly predicts all targets in one non-autoregressive pass. A single unified checkpoint retains STAMP's referring and reasoning capabilities while extending to open-vocabulary semantic, instance-aware, and remote-sensing small-target segmentation, where high-resolution mask-token scaling preserves finer spatial evidence. Across these settings, STAMPlus achieves state-of-the-art segmentation performance, preserves general multimodal instruction following, and reduces 12-category latency from 13.50s for repeated STAMP inference to 5.16s. Further analyses show that accurate target cues improve segmentation and learned spatial grounding benefits look-twice reasoning. Overall, STAMPlus resolves the trilemma beyond single-target prediction.
