PerceptionDLM: Parallel Region Perception with Multimodal Diffusion Language Models

2026-06-17

AI Summary

This work addresses the limitations of existing vision-language models in multi-region perception tasks, where autoregressive generation hinders parallel processing of descriptions for multiple image regions. To overcome this, we propose PerceptionDLM—the first multimodal parallel region-aware framework based on diffusion language modeling—which leverages structured attention masks and an efficient prompting mechanism to enable both sequence-level and token-level parallel description generation. We further introduce ParaDLC-Bench, a new benchmark designed to support parallel evaluation of such models. Experimental results demonstrate that PerceptionDLM achieves significantly improved inference efficiency while maintaining high description quality, exhibiting superior performance and scalability on ParaDLC-Bench.

Abstract

Multimodal large language models (MLLMs) have achieved remarkable progress in visual understanding tasks. However, most existing MLLMs rely on autoregressive generation, which limits their efficiency for perception tasks that require captioning multiple regions. In this work, we propose PerceptionDLM, a multimodal diffusion language model optimized for efficient parallel region perception. Built upon PerceptionDLM-Base, a strong foundational baseline that achieves state-of-the-art performance among open-source diffusion MLLMs, our architecture fully leverages the parallel decoding nature of DLMs. Specifically, we introduce efficient prompting and structured attention masking to enable simultaneous perception of multiple masked regions, allowing the model to generate region descriptions in parallel at both the sequence and token levels. This design significantly improves inference efficiency compared with existing approaches that process regions sequentially. To systematically evaluate the parallelism property of visual perception capability for DLMs, we construct a new Parallel Detailed Localized Captioning Benchmark (ParaDLC-Bench) by scaling the DLC-Bench to include multiple region masks per image, enabling joint evaluation of both caption quality and inference efficiency. Experiments demonstrate that PerceptionDLM maintains competitive performance in region captioning while achieving substantial speed improvements for multi-region perception tasks. Our results highlight the potential of multimodal diffusion language models for efficient, parallel visual perception. To the best of our knowledge, we are the first to achieve parallel region caption and perception by leveraging the advantages of diffusion language models. Code, models, and datasets are released.

Problem

• multimodal large language models
• parallel region perception
• region captioning
• visual understanding
• diffusion language models

Innovation

• diffusion language models
• parallel region perception
• multimodal large language models
• structured attention masking
• efficient visual captioning