CoMemo: LVLMs Need Image Context with Image Memory

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Shi, Su, Weijie, Zhu, Xizhou, Wang, Wenhai, Dai, Jifeng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908396712624128
author Liu, Shi
Su, Weijie
Zhu, Xizhou
Wang, Wenhai
Dai, Jifeng
author_facet Liu, Shi
Su, Weijie
Zhu, Xizhou
Wang, Wenhai
Dai, Jifeng
contents Recent advancements in Large Vision-Language Models built upon Large Language Models have established aligning visual features with LLM representations as the dominant paradigm. However, inherited LLM architectural designs introduce suboptimal characteristics for multimodal processing. First, LVLMs exhibit a bimodal distribution in attention allocation, leading to the progressive neglect of middle visual content as context expands. Second, conventional positional encoding schemes fail to preserve vital 2D structural relationships when processing dynamic high-resolution images. To address these limitations, we propose CoMemo - a dual-path architecture that combines a Context image path with an image Memory path for visual processing, effectively alleviating visual information neglect. Additionally, we introduce RoPE-DHR, a novel positional encoding mechanism that employs thumbnail-based positional aggregation to maintain 2D spatial awareness while mitigating remote decay in extended sequences. Evaluations across seven benchmarks,including long-context comprehension, multi-image reasoning, and visual question answering, demonstrate CoMemo's superior performance compared to conventional LVLM architectures. Project page is available at https://lalbj.github.io/projects/CoMemo/.
format Preprint
id arxiv_https___arxiv_org_abs_2506_06279
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CoMemo: LVLMs Need Image Context with Image Memory
Liu, Shi
Su, Weijie
Zhu, Xizhou
Wang, Wenhai
Dai, Jifeng
Computer Vision and Pattern Recognition
Recent advancements in Large Vision-Language Models built upon Large Language Models have established aligning visual features with LLM representations as the dominant paradigm. However, inherited LLM architectural designs introduce suboptimal characteristics for multimodal processing. First, LVLMs exhibit a bimodal distribution in attention allocation, leading to the progressive neglect of middle visual content as context expands. Second, conventional positional encoding schemes fail to preserve vital 2D structural relationships when processing dynamic high-resolution images. To address these limitations, we propose CoMemo - a dual-path architecture that combines a Context image path with an image Memory path for visual processing, effectively alleviating visual information neglect. Additionally, we introduce RoPE-DHR, a novel positional encoding mechanism that employs thumbnail-based positional aggregation to maintain 2D spatial awareness while mitigating remote decay in extended sequences. Evaluations across seven benchmarks,including long-context comprehension, multi-image reasoning, and visual question answering, demonstrate CoMemo's superior performance compared to conventional LVLM architectures. Project page is available at https://lalbj.github.io/projects/CoMemo/.
title CoMemo: LVLMs Need Image Context with Image Memory
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.06279