CORTEXA
← Browse
arxivcs.CVcs.AI2026-07-08

AnchorPrune: Relevance-Anchored Contextual Expansion for Visual Token Pruning

Kyuan Oh, Bumsoo Kim

Large vision-language models incur substantial inference costs because high-resolution inputs introduce thousands of visual tokens, many of which are redundant for a given query. Existing pruning methods often combine query relevance and token diversity, yet these objectives can conflict under aggressive compression: relevance-driven selection may overconcentrate the budget on correlated local evidence, while diversity-driven selection may suppress indispensable tokens or retain distinct but uninformative regions. We introduce AnchorPrune, a training-free framework that first constructs a protected relevance anchor and then expands it with complementary visual context. AnchorPrune adaptively determines the anchor size from the novelty profile of relevance-ranked tokens, preserving a compact set of query-critical evidence, and allocates the remaining budget through importance-weighted novelty to recover informative, non-redundant context relative to the anchor. This ordered design prevents contextual expansion from displacing indispensable query cues while improving overall visual coverage. AnchorPrune is lightweight, architecture-aware, and requires neither retraining nor model modification. Across image and video vision-language models and benchmarks, it consistently improves the accuracy-efficiency trade-off over training-free baselines, particularly under severe compression. On LLaVA-NeXT-7B, AnchorPrune preserves 97.6% of full-token performance using only 160 of 2,880 visual tokens. These results establish relevance-anchored contextual expansion as an effective principle for efficient multimodal inference. Code is available at https://github.com/MULTI-cau/AnchorPrune.

View free PDFSource page

Related papers

arxivcs.CVcs.AI2026-07-09

VocaDet: Sample-Driven Open-Vocabulary Object Detection and Segmentation via Visual Tokenization and Vector Database Retrieval

ZhiXin Sun

Open-vocabulary object detection and segmentation aim to recognize arbitrary objects beyond predefined categories. Although recent vision-language and reference-based approaches have significantly advanced this field, they often rely on text prompts, limited visual examples, or e…

View free PDFSource page
arxivcs.CVcs.AI2026-07-05

SeeMe: Mitigating Hallucinations in Large Vision-Language Models through Effective Visual Token Engineering

Kai Tang, Jinhao You, Bohua Zhang, Yichen Guo, Yiding Sun, Dongxu Zhang, et al.

Large Vision-Language Models (LVLMs) have achieved remarkable progress in visual understanding tasks such as image captioning and visual question answering. However, they remain susceptible to hallucinations, generating content that is inconsistent with the actual visual input. E…

View free PDFSource page
arxivcs.IRcs.AIcs.CLcs.CV2026-07-06

Do All Visual Tokens Matter Equally? Object-Evidence Preserving Token Merging for Vision-Language Retrieval

Suhyeong Park, Junha Jung, Jungwoo Park, Jaewoo Kang

Multi-vector vision-language retrieval preserves fine-grained visual evidence through maximum-similarity late interaction, but dense image-side tokens make storage and scoring expensive. Existing token compression methods reduce this cost, yet they can remove or collapse object-…

View free PDFSource page
arxivcs.CVcs.AIcs.LG2026-07-05

Seeing Once is Enough? Online Geometry-Aware Token Pruning for 3D Question Answering

Ruei-Chi Lai, Bolivar Solarte, Chin-Hsuan Wu, Yi-Hsuan Tsai, Min Sun

Recent Multi-modal Large Language Models (MLLMs) have demonstrated remarkable performance on 2D question answering tasks. However, extending these models to the 3D question answering remains challenging, as they typically require multiple views of the scene, which incurs substant…

View free PDFSource page
arxivcs.CVcs.AI2026-06-28

Fast Enough to Act: Spatio-Temporal Visual Token Merging for Low-Latency Robotic VLMs and VLAs

Junzhou Chen, Jindong Wang, Gang Zhou

Vision-language models and vision-language action models endow the robot with unprecedented capabilities. However, the input of video and high-resolution images yields a massive number of visual tokens, leading to extremely high inference latency and severely hindering the robot'…

View free PDFSource page