Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models

• URL: http://arxiv.org/abs/2605.14530v2
• Date: Tue, 19 May 2026 00:58:43 GMT
• Title: Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models
• Abstract summary: We show that LDVLMs suffer from repetitive generation and degraded visual grounding.<n>We propose a training-free approach, introducing Mask Prior Suppression and Monotonic RoPE Scaling.<n>Our results show that these failures can be effectively addressed with a lightweight, plug-and-play strategy.
• Score: 7.964052580720558
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large diffusion vision-language models (LDVLMs) have recently emerged as a promising alternative to autoregressive models, enabling parallel decoding for efficient inference and leveraging bidirectional attention for global context. Despite these advances, their behavior under long-form generation remains underexplored. In this work, we show that existing LDVLMs suffer from repetitive generation and degraded visual grounding, and identify two underlying causes. First, repetitive generation originates from a mask token prior: since generation tokens are initialized as mask tokens, their hidden representations progressively drift toward a shared prior direction over generation steps. Second, a fundamental misalignment between the positional attention bias and the iterative unmasking process suppresses attention toward informative visual tokens, degrading visual grounding. Based on these insights, we propose a training-free approach, introducing Mask Prior Suppression and Monotonic RoPE Scaling to mitigate mask prior drift and positional attention collapse during decoding. Experiments on general multimodal benchmarks and visual grounding tasks demonstrate improvements over baseline LDVLMs, with robust gains on long-form description benchmarks. Our results show that these failures can be effectively addressed with a lightweight, plug-and-play strategy that requires no additional training and generalizes across diverse LDVLM architectures.

Related papers

• Nemotron-Labs-Diffusion-Image: Advancing Masked Discrete Diffusion for High-Resolution Image Synthesis [93.7120829382642]
  Nemotron-Labs-Diffusion-Image is a state-of-the-art masked discrete diffusion model for text-to-image synthesis.<n>Standard MDMs lack self-correcting capability because discrete tokens cannot be modified once they are unmasked.<n>Nemotron-Labs-Diffusion-Image incorporates a token-editing mechanism that enables the model to dynamically revise already-unmasked tokens during inference.
  arXiv  Detail & Related papers  (2026-06-29T05:48:41Z)
• Masked Diffusion Vision-Language Models for Temporal Action Localization [10.018515245238216]
  Temporal action localization (TAL) requires recognizing the target event and localizing its start and end times precisely in untrimmed videos.<n>Recent vision-language formulations improve semantic reasoning and support language-conditioned outputs, but their autoregressive decoders still generate tokens from left to right.<n>We adapt vision-language models to TAL so that semantic tokens and boundary tokens remain editable throughout iterative denoising.
  arXiv  Detail & Related papers  (2026-05-28T12:39:04Z)
• Sketch Then Paint: Hierarchical Reinforcement Learning for Diffusion Multi-Modal Large Language Models [52.40742159500277]
  We show how to optimize Diffusion Multi-Modal Large Language Models (dMLLMs) through reinforcement learning (RL)<n>Our approach features a Sketch-Then-Paint training scheme that organizes updates into three distinct stages: global, structure, and refinement.<n> Experiments using two popular dMLLM backbones, MMaDA and Lumina-DiMOO, demonstrate substantial gains on the GenEval and DPG benchmarks.
  arXiv  Detail & Related papers  (2026-05-16T06:59:54Z)
• Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation [51.743225614196774]
  Multimodal large language models (MLLMs) have achieved remarkable progress in vision-language reasoning.<n>They remain vulnerable to hallucination, where generated content deviates from visual evidence.<n>Recent vision enhancement methods attempt to address this issue by reinforcing visual tokens during decoding.<n>We propose Adaptive Visual Reinforcement (AIR), a training-free framework for MLLMs.
  arXiv  Detail & Related papers  (2026-02-27T14:18:51Z)
• WeMMU: Enhanced Bridging of Vision-Language Models and Diffusion Models via Noisy Query Tokens [69.97021957331326]
  We propose Noisy Query Tokens, which learn a distributed representation space between the VLM and Diffusion Model via end-to-end optimization.<n>We also introduce a VAE branch with linear projection to recover fine-grained image details.
  arXiv  Detail & Related papers  (2025-12-02T09:02:20Z)
• Masks Can Be Distracting: On Context Comprehension in Diffusion Language Models [19.847438086389616]
  Masked Diffusion Language Models have emerged as a promising alternative to Autoregressive Language Models.<n>We show that MDLMs exhibit a strong locality bias, favouring local over distant context.<n>We introduce a mask-agnostic loss function that encourages predictions to remain invariant to the number of appended masks.
  arXiv  Detail & Related papers  (2025-11-26T12:44:29Z)
• A Comprehensive Study on Visual Token Redundancy for Discrete Diffusion-based Multimodal Large Language Models [85.30893355216486]
  We study how visual token redundancy evolves with different dMLLM architectures and tasks.<n>Our study reveals that visual redundancy emerges only in from-scratch dMLLMs while handling long-answer tasks.<n>Layer-skipping is promising for accelerating AR-to-diffusion dMLLMs, whereas progressive or late-step pruning is more effective for from-scratch dMLLMs.
  arXiv  Detail & Related papers  (2025-11-19T04:13:36Z)
• Multimodal LLM-Guided Semantic Correction in Text-to-Image Diffusion [52.315729095824906]
  MLLM Semantic-Corrected Ping-Pong-Ahead Diffusion (PPAD) is a novel framework that introduces a Multimodal Large Language Model (MLLM) as a semantic observer during inference.<n>It performs real-time analysis on intermediate generations, identifies latent semantic inconsistencies, and translates feedback into controllable signals that actively guide the remaining denoising steps.<n>Extensive experiments demonstrate PPAD's significant improvements.
  arXiv  Detail & Related papers  (2025-05-26T14:42:35Z)
• Efficient Multi-modal Large Language Models via Visual Token Grouping [55.482198808206284]
  High-resolution images and videos pose a barrier to their broader adoption.<n> compressing vision tokens in MLLMs has emerged as a promising approach to reduce inference costs.<n>We introduce VisToG, a novel grouping mechanism that leverages the capabilities of pre-trained vision encoders to group similar image segments.
  arXiv  Detail & Related papers  (2024-11-26T09:36:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.