ALTo: Adaptive-Length Tokenizer for Autoregressive Mask Generation
This addresses the efficiency and quality trade-off in segmentation tasks for multimodal AI systems, though it appears incremental as it builds on existing tokenization and optimization methods.
The paper tackles the problem of rigid token representations in multimodal large language models for segmentation by proposing ALTo, an adaptive-length tokenizer for autoregressive mask generation, achieving state-of-the-art performance with adaptive token cost on popular benchmarks.
While humans effortlessly draw visual objects and shapes by adaptively allocating attention based on their complexity, existing multimodal large language models (MLLMs) remain constrained by rigid token representations. Bridging this gap, we propose ALTo, an adaptive length tokenizer for autoregressive mask generation. To achieve this, a novel token length predictor is designed, along with a length regularization term and a differentiable token chunking strategy. We further build ALToLLM that seamlessly integrates ALTo into MLLM. Preferences on the trade-offs between mask quality and efficiency is implemented by group relative policy optimization (GRPO). Experiments demonstrate that ALToLLM achieves state-of-the-art performance with adaptive token cost on popular segmentation benchmarks. Code and models are released at https://github.com/yayafengzi/ALToLLM.