NewEvery arXiv paper, its researchers & institutions — mapped.
computer vision

ReToken: One Token to Improve Vision-Language Models for Visual Retrieval

arXiv:2607.28627

summary

ReToken introduces a single learnable embedding that acts as a retrieval token to select a sparse set of relevant visual tokens from a cached representation, improving vision-language model performance on long visual contexts and visual retrieval tasks.

Abstract

Long visual context poses a challenge for vision-language models: performance degrades as the number of distractors grows, and processing all tokens at once is computationally infeasible under GPU memory constraints. We present ReToken, a single learnable embedding trained as an explicit retrieval target that selects a sparse set of query-relevant visual tokens from a pre-filled visual KV cache. Trained on only a small image-QA dataset, ReToken yields consistent gains across image and video benchmarks: on Visual Haystacks it improves Qwen3VL-8B by 13.4 points and InternVL3.5 by 12.4 points (>20% relative), and on LVBench it transfers zero-shot to long video for an 8.0-point gain with Qwen3VL-8B. Thanks to its lightweight design, both training and long-video inference fit on a single H100. Code is available at: https://github.com/avaxiao/ReToken

Code: https://github.com/avaxiao/ReToken

Topics & keywords

#vision-language models#visual retrieval#sparse token selection#long video processing#multimodal retrievallearnable embeddingvisual KV cacheretrieval tokenQwen3VLInternVLLVBench