Decoupled Visual Processing: Efficient Multimodal Adaptation via Modality-Specific Transformer Substitution

• URL: http://arxiv.org/abs/2607.26596v1
• Date: Wed, 29 Jul 2026 08:16:28 GMT
• Title: Decoupled Visual Processing: Efficient Multimodal Adaptation via Modality-Specific Transformer Substitution
• Abstract summary: We propose Decoupled Visual Processing (DVP) as an efficient training framework for multimodal large language models.<n>During training, only the single transformer block is updated, dramatically reducing the number of trainable parameters.<n>Experiments on the LLaVA-1.5 framework demonstrate that DVP competitive performance on MME, POPE, and ChartQA benchmarks while training only a fraction of the total parameters.
• Score: 26.422874365108985
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multimodal large language models (MLLMs) have demonstrated remarkable capabilities by integrating visual and textual understanding within a unified transformer architecture. However, fine-tuning all parameters of these models for visual instruction tuning is computationally expensive and often unnecessary, as the representation requirements for visual and textual tokens diverge significantly in the deeper layers of the network. In this paper, we propose Decoupled Visual Processing (DVP), an efficient training framework that replaces the upper decoder layers of a pretrained LLM with a lightweight, independently trainable single transformer block dedicated exclusively to visual token processing. Specifically, after shared processing through the first half of the decoder layers, visual and textual tokens are split: visual tokens are routed through a newly initialized single transformer block while textual tokens continue through the original frozen decoder layers. The two streams are then concatenated before the language modeling head. During training, only the single transformer block is updated, dramatically reducing the number of trainable parameters. Experiments on the LLaVA-1.5 framework demonstrate that DVP achieves competitive performance on MME, POPE, and ChartQA benchmarks while training only a fraction of the total parameters, suggesting that visual representations in MLLMs can be effectively learned through a decoupled, parameter-efficient pathway.

Related papers

• Let ViT Speak: Generative Language-Image Pre-training [102.76320739909421]
  GenLIP is a minimalist generative pretraining framework for Vision Transformers (ViTs)<n>It trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective.<n>It achieves competitive or superior results across diverse multimodal benchmarks.
  arXiv  Detail & Related papers  (2026-05-01T17:51:38Z)
• Patch-as-Decodable-Token: Towards Unified Multi-Modal Vision Tasks in MLLMs [88.68484904214142]
  We introduce Patch-as-Decodable Token (PaDT) to generate both textual and diverse visual outputs.<n>Central to PaDT are Visual Reference Tokens (VRTs), derived from visual patch embeddings of query images.<n>We show PaDT consistently achieves state-of-the-art performance, even compared with significantly larger MLLM models.
  arXiv  Detail & Related papers  (2025-10-02T12:23:57Z)
• VoxelFormer: Parameter-Efficient Multi-Subject Visual Decoding from fMRI [4.3296865400748]
  VoxelFormer is a lightweight transformer architecture that enables multi-subject training for visual decoding from fMRI.<n>It integrates a Token Merging Transformer (ToMer) for efficient voxel compression and a query-driven Q-Former that produces fixed-size neural representations aligned with the CLIP image embedding space.
  arXiv  Detail & Related papers  (2025-09-10T21:20:17Z)
• CoViPAL: Layer-wise Contextualized Visual Token Pruning for Large Vision-Language Models [75.88232735646018]
  Large Vision-Language Models (LVLMs) process multimodal inputs consisting of text tokens and vision tokens extracted from images or videos.<n>Existing methods attempt to prune redundant vision tokens, revealing substantial redundancy in visual representations.<n>We propose CoViPAL, a layer-wise contextualized visual token pruning method that employs a Plug-and-Play Pruning Module (PPM) to predict and remove redundant vision tokens before they are processed by the LVLM.
  arXiv  Detail & Related papers  (2025-08-24T07:47:00Z)
• DyMU: Dynamic Merging and Virtual Unmerging for Efficient VLMs [124.52164183968145]
  We present DyMU, an efficient, training-free framework that reduces the computational burden of vision-language models (VLMs)<n>Our approach comprises two key components. First, Dynamic Token Merging (DToMe) reduces the number of visual token embeddings by merging similar tokens based on image complexity.<n>Second, Virtual Token Unmerging (VTU) simulates the expected token sequence for large language models (LLMs) by efficiently reconstructing the attention dynamics of a full sequence.
  arXiv  Detail & Related papers  (2025-04-23T18:38:18Z)
• Sparsity Meets Similarity: Leveraging Long-Tail Distribution for Dynamic Optimized Token Representation in Multimodal Large Language Models [6.467840081978855]
  multimodal large language models (MM-LLMs) have achieved significant success in various tasks.<n>Main computational burden arises from processingd text and visual tokens.<n>We propose a dynamic pruning algorithm that identifies the inflection point in the visual CLS token similarity curve.
  arXiv  Detail & Related papers  (2024-09-02T10:49:10Z)
• Enhancing Vision-Language Model with Unmasked Token Alignment [37.12838142681491]
  This paper introduces Unmasked Token Alignment (UTA), a method that leverages existing CLIP models to further enhance its vision-language representations. UTA trains a Vision Transformer (ViT) by aligning unmasked visual tokens to the corresponding image tokens from a frozen CLIP vision encoder, which automatically aligns the ViT model with the CLIP text encoder.
  arXiv  Detail & Related papers  (2024-05-29T11:48:17Z)
• TVLT: Textless Vision-Language Transformer [89.31422264408002]
  We present the Textless Vision-Language Transformer (TVLT), where homogeneous transformer blocks take raw visual and audio inputs. TVLT attains performance comparable to its text-based counterpart, on various multimodal tasks. Our findings suggest the possibility of learning compact and efficient visual-linguistic representations from low-level visual and audio signals.
  arXiv  Detail & Related papers  (2022-09-28T15:08:03Z)
• Learning Visual Representation from Modality-Shared Contrastive Language-Image Pre-training [88.80694147730883]
  We investigate a variety of Modality-Shared Contrastive Language-Image Pre-training (MS-CLIP) frameworks. In studied conditions, we observe that a mostly unified encoder for vision and language signals outperforms all other variations that separate more parameters. Our approach outperforms vanilla CLIP by 1.6 points in linear probing on a collection of 24 downstream vision tasks.
  arXiv  Detail & Related papers  (2022-07-26T05:19:16Z)
• All in One: Exploring Unified Video-Language Pre-training [44.22059872694995]
  We introduce an end-to-end video-language model, namely textitall-in-one Transformer, that embeds raw video and textual signals into joint representations. The code and pretrained model have been released in https://github.com/showlab/all-in-one.
  arXiv  Detail & Related papers  (2022-03-14T17:06:30Z)

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.