ConsensusDrop: Fusing Visual and Cross-Modal Saliency for Efficient Vision Language Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Parikh, Dhruv, Fan, Haoyang, Kannan, Rajgopal, Prasanna, Viktor
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866918317788233728
author Parikh, Dhruv
Fan, Haoyang
Kannan, Rajgopal
Prasanna, Viktor
author_facet Parikh, Dhruv
Fan, Haoyang
Kannan, Rajgopal
Prasanna, Viktor
contents Vision-Language Models (VLMs) are expensive because the LLM processes hundreds of largely redundant visual tokens. Existing token reduction methods typically exploit \textit{either} vision-encoder saliency (broad but query-agnostic) \textit{or} LLM cross-attention (query-aware but sparse and costly). We show that neither signal alone is sufficient: fusing them consistently improves performance compared to unimodal visual token selection (ranking). However, making such fusion practical is non-trivial: cross-modal saliency is usually only available \emph{inside} the LLM (too late for efficient pre-LLM pruning), and the two signals are inherently asymmetric, so naive fusion underutilizes their complementary strengths. We propose \textbf{ConsensusDrop}, a training-free framework that derives a \emph{consensus} ranking by reconciling vision encoder saliency with query-aware cross-attention, retaining the most informative tokens while compressing the remainder via encoder-guided token merging. Across LLaVA-1.5/NeXT, Video-LLaVA, and other open-source VLMs, ConsensusDrop consistently outperforms prior pruning methods under identical token budgets and delivers a stronger accuracy-efficiency Pareto frontier -- preserving near-baseline accuracy even at aggressive token reductions while reducing TTFT and KV cache footprint. Our code will be open-sourced.
format Preprint
id arxiv_https___arxiv_org_abs_2602_00946
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ConsensusDrop: Fusing Visual and Cross-Modal Saliency for Efficient Vision Language Models
Parikh, Dhruv
Fan, Haoyang
Kannan, Rajgopal
Prasanna, Viktor
Computer Vision and Pattern Recognition
Vision-Language Models (VLMs) are expensive because the LLM processes hundreds of largely redundant visual tokens. Existing token reduction methods typically exploit \textit{either} vision-encoder saliency (broad but query-agnostic) \textit{or} LLM cross-attention (query-aware but sparse and costly). We show that neither signal alone is sufficient: fusing them consistently improves performance compared to unimodal visual token selection (ranking). However, making such fusion practical is non-trivial: cross-modal saliency is usually only available \emph{inside} the LLM (too late for efficient pre-LLM pruning), and the two signals are inherently asymmetric, so naive fusion underutilizes their complementary strengths. We propose \textbf{ConsensusDrop}, a training-free framework that derives a \emph{consensus} ranking by reconciling vision encoder saliency with query-aware cross-attention, retaining the most informative tokens while compressing the remainder via encoder-guided token merging. Across LLaVA-1.5/NeXT, Video-LLaVA, and other open-source VLMs, ConsensusDrop consistently outperforms prior pruning methods under identical token budgets and delivers a stronger accuracy-efficiency Pareto frontier -- preserving near-baseline accuracy even at aggressive token reductions while reducing TTFT and KV cache footprint. Our code will be open-sourced.
title ConsensusDrop: Fusing Visual and Cross-Modal Saliency for Efficient Vision Language Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.00946