CVPR 2026 | LLM × Graph论文总结（VLM，多模态大模型，问答，Graph4VLM等）

CVPR 2026将在2026年6月5日至7日于美国科罗拉多会议中心（Colorado Convention Center）举行。

本文总结了CVPR 2026上有关LLM × Graph的相关论文。

LLM × Graph Topic：VLM，多模态大模型，问答，Graph4VLM等。

CVPR2026论文列表：https://cvpr.thecvf.com/virtual/2026/papers.html

点击文末阅读原文跳转笔者知乎链接（跳转论文链接更方便）。

1 Beyond Graph Model: Reliable VLM Fine-Tuning via Random Graph Adapter

链接：https://cvpr.thecvf.com/virtual/2026/poster/39164

arXiv：https://arxiv.org/abs/2507.10355

作者：Bo Jiang, Xueyang Ze, Beibei Wang, Xixi Wang, Xixi Wan, Bin Luo

关键词：VLM，图适应器

2 DynamicGTR: Leveraging Graph Topology Representation Preferences to Boost VLM Capabilities on Graph QAs

链接：https://cvpr.thecvf.com/virtual/2026/poster/39345

arXiv：http://arxiv.org/abs/2602.21864

作者：Yanbin Wei, Jiangyue Yan, Chun Kang, Yang Chen, Hua Liu, James Kwok, Yu Zhang

关键词：图问答，VLM，拓扑表示

3 Training-Only Heterogeneous Image-Patch-Text Graph Supervision for Advancing Few-Shot Learning Adapters

链接：https://cvpr.thecvf.com/virtual/2026/poster/39034

arXiv：http://arxiv.org/abs/2603.18101

代码：https://github.com/MR-Sherif/TOGA

作者：Mohammed Rahman Sherif Khan Mohammad, Ardhendu Behera, Sandip Pradhan, Swagat Kumar, Amr Ahmed

关键词：小样本,VLM，异构图

4 Mario: Multimodal Graph Reasoning with Large Language Models

链接：https://cvpr.thecvf.com/virtual/2026/poster/40048

arXiv：http://arxiv.org/abs/2603.05181

代码：https://github.com/sunyuanfu/Mario

作者：Yuanfu Sun, Kang Li, Pengkang Guo, Jiajin Liu, Qiaoyu Tan

关键词：多模态图推理，LLM

5 Structural Graph Probing of Vision-Language Models

链接：https://cvpr.thecvf.com/virtual/2026/poster/40160

arXiv：https://arxiv.org/abs/2603.27070

作者：Haoyu He, Yue Zhuo, Yu Zheng, Qi Wang

关键词：VLM，神经拓扑，GNN

6 DiGraphHal-Bench: Evaluating Multimodal Large Language Models on Complex Directed Graphs

链接：https://cvpr.thecvf.com/virtual/2026/poster/39910

作者：Yixin Fan, He Zhao, Yuxin Hou, Changhua Zhou, Zihao Liu, Peng Wang, Lu ChengLong, Xu Zhang, Wei Wang

关键词：视觉QA，benchmark，图理解

7 GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning

链接：https://cvpr.thecvf.com/virtual/2026/poster/37657

arXiv：http://arxiv.org/abs/2603.13370

代码：https://github.com/oamyjin/GraphVLM

作者：Jiajin Liu, Dongzhe Fan, Chuanhao Ji, Daochen Zha, Qiaoyu Tan

关键词：benchmark，VLM，多模态图学习

8 CASPA: Graph-Structured Concept Anchors for Modality-Agnostic Adaptation in Vision-Language Models

链接：https://cvpr.thecvf.com/virtual/2026/poster/39522

作者：Abhiroop Chatterjee, Susmita Ghosh, Ashish Ghosh, Emmett Ientilucci

关键词：图结构，模型无关适应，VLM