DAVET: Denoising-Aware Visual Evidence Trajectory Allocation for Diffusion Vision-Language Models

• URL: http://arxiv.org/abs/2608.01821v1
• Date: Mon, 03 Aug 2026 07:29:05 GMT
• Title: DAVET: Denoising-Aware Visual Evidence Trajectory Allocation for Diffusion Vision-Language Models
• Abstract summary: Diffusion vision-language models (dVLMs) iteratively denoise masked responses while conditioning each denoising step on visual evidence.<n>We present Denoising-Aware Visual Evidence Trajectory Allocation (DAVET), a training-free framework that allocates visual evidence according to the evolving generation state.
• Score: 5.2649891485209634
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Diffusion vision-language models (dVLMs) iteratively denoise masked responses while conditioning each denoising step on visual evidence, making visual conditioning a substantial recurring inference cost. Unlike autoregressive decoding, diffusion generation repeatedly revisits the entire response as uncertainty evolves. Our analysis reveals that visual evidence demand is strongly step-dependent, motivating adaptive allocation across denoising steps. Existing inference acceleration methods operate through decoding-side strategies or visual token compression via pruning and merging, but do not explicitly treat visual evidence as a resource whose demand evolves across the diffusion process. Therefore, we present Denoising-Aware Visual Evidence Trajectory Allocation (DAVET), a training-free framework that allocates visual evidence according to the evolving generation state. Starting from a phase-conditioned evidence trajectory, the proposed allocation policy uses operation demand to set an evidence reserve whose allocation at each denoising step is modulated by trajectory risk. DAVET realizes the resulting budgets through a hierarchy of evidence views constructed from a single visual encoding, separating when and how much evidence is needed from how the evidence views are constructed. Evaluated on two representative dVLMs, LLaDA-V and LaViDa, across multiple visual-understanding benchmarks, DAVET achieves an average speedup of 1.55$\times$ with an average relative performance drop of 1.86\%, showing that denoising-aware visual evidence allocation can reduce visual conditioning cost while largely preserving generation quality.

Related papers

• Visual Representation Matters: Exploiting Temporal Differences in Video-to-Audio Generation [53.47090840868696]
  Video-to-audio (V2A) generation extends image-to-audio generation (I2A)<n>We introduce TD-V2A, which leverages temporal differences (TD) as the key representation that distinguishes V2A from I2A.<n>We show that effectively exploiting TD through our proposed framework significantly improves end-to-end V2A generation quality.
  arXiv  Detail & Related papers  (2026-08-05T14:28:05Z)
• Probing Diffusion Denoising Dynamics for Contrastive Representation Learning [48.679227622825806]
  We study how can the denoising dynamics of a pretrained diffusion model be adapted to support discriminative representation learning.<n>We present D$3$CL as an investigation of this question.<n>We show that noise-level contrastive learning can structure denoising representations for discriminative tasks while maintaining generative performance.
  arXiv  Detail & Related papers  (2026-07-10T03:23:44Z)
• Persistent Visual Memory: Sustaining Perception for Deep Generation in LVLMs [66.15429821530503]
  Persistent Visual Memory is a lightweight learnable module designed to strengthen sustained, on-demand access to visual evidence.<n>Experiments on Qwen3-VL models demonstrate that PVM brings notable improvements with negligible parameter overhead.<n>In-depth analysis reveals that PVM shows improved robustness in longer generations and accelerates internal prediction convergence.
  arXiv  Detail & Related papers  (2026-05-01T17:54:37Z)
• Instruction-Evidence Contrastive Dual-Stream Decoding for Grounded Vision-Language Reasoning [3.944710336786154]
  Vision-Language Models (VLMs) exhibit strong performance in instruction following and open-ended vision-language reasoning.<n>They frequently generate fluent outputs that are weakly grounded in visual evidence.<n>We propose a decoding framework that balances linguistic informativeness and visual faithfulness during generation.
  arXiv  Detail & Related papers  (2026-04-28T16:18:31Z)
• Mutual Learning for Acoustic Matching and Dereverberation via Visual Scene-driven Diffusion [93.32354378820648]
  We introduce MVSD, a mutual learning framework based on diffusion models. MVSD considers the two tasks symmetrically, exploiting the reciprocal relationship to facilitate learning from inverse tasks. Our framework can improve the performance of the reverberator and dereverberator.
  arXiv  Detail & Related papers  (2024-07-15T00:47:56Z)
• DiffusionVMR: Diffusion Model for Joint Video Moment Retrieval and Highlight Detection [38.12212015133935]
  A novel framework, DiffusionVMR, is proposed to redefine the two tasks as a unified conditional denoising generation process. Experiments conducted on five widely-used benchmarks demonstrate the effectiveness and flexibility of the proposed DiffusionVMR.
  arXiv  Detail & Related papers  (2023-08-29T08:20:23Z)
• DiffSED: Sound Event Detection with Denoising Diffusion [70.18051526555512]
  We reformulate the SED problem by taking a generative learning perspective. Specifically, we aim to generate sound temporal boundaries from noisy proposals in a denoising diffusion process. During training, our model learns to reverse the noising process by converting noisy latent queries to the groundtruth versions.
  arXiv  Detail & Related papers  (2023-08-14T17:29:41Z)
• Denoising-Contrastive Alignment for Continuous Sign Language Recognition [22.800767994061175]
  Continuous sign language recognition aims to recognize signs in untrimmed sign language videos to textual glosses.<n>Current cross-modality alignment paradigms often neglect the role of textual grammar to guide the video representation.<n>We propose a Denoising-Contrastive Alignment paradigm to enhance video representations.
  arXiv  Detail & Related papers  (2023-05-05T15:20:27Z)
• Fully Unsupervised Diversity Denoising with Convolutional Variational Autoencoders [81.30960319178725]
  We propose DivNoising, a denoising approach based on fully convolutional variational autoencoders (VAEs) First we introduce a principled way of formulating the unsupervised denoising problem within the VAE framework by explicitly incorporating imaging noise models into the decoder. We show that such a noise model can either be measured, bootstrapped from noisy data, or co-learned during training.
  arXiv  Detail & Related papers  (2020-06-10T21:28:13Z)

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.