SAMTok: Representing Any Mask with Two Words

2026-01-22

AI Summary

This work addresses the challenge of enabling multimodal large language models (MLLMs) to efficiently support pixel-level tasks, which has been hindered by complex region encoders, task-specific decoders, and incompatible training objectives. The authors propose SAMTok, a novel approach that compresses arbitrary region masks into just two discrete language tokens, allowing standard MLLMs to learn pixel-level understanding and generation through conventional language modeling—without architectural modifications or custom loss functions. Built upon SAM2, SAMTok integrates a mask encoder with a residual vector quantizer and is trained on 209 million masks, further refined via reinforcement learning with a text-mask alignment reward. An implementation based on Qwen-VL achieves state-of-the-art or competitive performance on region captioning, region-based VQA, and referring segmentation, while significantly improving interactive segmentation results on the GRES and GCG benchmarks.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Computer Vision: Large Vision ModelsNatural Language Processing: Language Grounding & Multi-modal NLP

Application Category

Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved informationUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationEconomics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Pixel-wise capabilities are essential for building interactive intelligent systems. However, pixel-wise multi-modal LLMs (MLLMs) remain difficult to scale due to complex region-level encoders, specialized segmentation decoders, and incompatible training objectives. To address these challenges, we present SAMTok, a discrete mask tokenizer that converts any region mask into two special tokens and reconstructs the mask using these tokens with high fidelity. By treating masks as new language tokens, SAMTok enables base MLLMs (such as the QwenVL series) to learn pixel-wise capabilities through standard next-token prediction and simple reinforcement learning, without architectural modifications and specialized loss design. SAMTok builds on SAM2 and is trained on 209M diverse masks using a mask encoder and residual vector quantizer to produce discrete, compact, and information-rich tokens. With 5M SAMTok-formatted mask understanding and generation data samples, QwenVL-SAMTok attains state-of-the-art or comparable results on region captioning, region VQA, grounded conversation, referring segmentation, scene graph parsing, and multi-round interactive segmentation. We further introduce a textual answer-matching reward that enables efficient reinforcement learning for mask generation, delivering substantial improvements on GRES and GCG benchmarks. Our results demonstrate a scalable and straightforward paradigm for equipping MLLMs with strong pixel-wise capabilities. Our code and models are available.

Problem

• pixel-wise capabilities
• multimodal LLMs
• mask representation
• scalable segmentation
• interactive intelligent systems

Innovation

• mask tokenization
• pixel-wise MLLM
• discrete representation
• reinforcement learning
• next-token prediction