ACM MM - arXivDaily 期刊&会议

2608.03559 2026-08-05 cs.CV 新提交

Compass: Degradation-Simulated Reciprocal Learning with Lightweight Needle RWKV for Multimodal Crack Segmentation under Missing Modalities

Compass：面向缺失模态场景下多模态裂缝分割的退化模拟互学习与轻量型Needle RWKV

Hui Liu, Chen Jia, Fan Shi, Xu Cheng, Mianzhao Wang, Shengyong Chen

AI总结 Compass是面向缺失模态场景的多模态裂缝分割轻量型网络，通过DSD、Needle Block与ETPF实现鲁棒分割，在90%深度模态缺失时仍达SOTA性能，仅需258万参数。

Comments This paper has been accepted by ACM MM 2026

2608.03450 2026-08-05 cs.MM cs.AI cs.CL cs.CV 新提交

Balancing Efficiency and Efficacy: Training-Free Attention-Guided Switching Between Explicit and Latent Thoughts for MLLMs

平衡效率与效能：面向多模态大语言模型（MLLM）的无训练注意力引导式显式与隐式思维切换

Haoqian Kang, Liupeng Li, Kuofeng Gao, Jinpeng Wang, Zhenyu Lu, Bin Chen, Ke Chen, Yaowei Wang

AI总结针对MLLM推理中感知与逻辑混淆的问题，提出无训练的注意力引导式切换框架，通过视觉-文本注意力比率自适应切换显式与隐式推理，实现性能与效率的双重提升。

Comments Accepted by ACM MM 2026. 10 pages, 6 figures, 5 tables

2608.03264 2026-08-05 cs.MM cs.CV cs.SD 新提交

Hear to See: Discerning Stateful Listening for Audio-Visual Instance Segmentation

听而能视：面向视听实例分割的状态感知聆听

Leiye Liu, Miao Zhang, Jiahong Jiang, Jingjing Li, Jialong Zhong, Kai Peng, Tingwei Liu, Wei Ji, Yongri Piao, Huchuan Lu

AI总结本文提出 Hear to See（H2S）模型，通过 ASP 和 ADM 机制解决视听实例分割中重叠声源匹配与异步动态跟踪问题，在 AVISeg 数据集上实现 SOTA 性能，mAP 达 48.54，较此前方法提升 7.8%。

Comments Accepted by ACM MM 2026

2608.03225 2026-08-05 cs.CV 新提交

Open-Linguistic Concept Unified Learning for Cross-Site Interpretable Dermatology Image Diagnosis

面向跨站点可解释皮肤病图像诊断的开放语言概念统一学习

Chengyu Wu, Junpeng Tan, Wanxiang Luo, Yaqi Wang, Yandong Wen, Yefeng Zheng

AI总结针对现有概念模型跨站点泛化差、适配FVLMs成本高的问题，提出开放语言概念统一学习框架UniCon，通过共享语义空间等三项创新实现多模态可解释皮肤病诊断，获顶级准确率且具备跨站点干预能力。

Comments accepted by ACM Multimedia 2026

2608.03151 2026-08-05 cs.CR cs.HC 新提交

AirKey: Multimodal Acoustic-Assisted WiFi Sensing for Zero-Training Robust PIN Inference

AirKey：用于零训练鲁棒PIN推断的多模态声学辅助WiFi感知

BaiChuan Wu, Bin Liu, Xiang Zhang, Zhi Liu, Jie Zhang, Chao Liu, Huan Yan, Meng Li, Fusang Zhang

AI总结 AirKey是一种多模态感知框架，通过利用IEEE 802.11机制和声学信号辅助WiFi感知，实现零训练鲁棒PIN推断，准确率超现有单模态方案4倍，可在6次尝试内恢复设备解锁PIN，凸显智能界面的隐私漏洞。

Comments Accepted by ACM MM 2026

2608.03120 2026-08-05 cs.CV 新提交

SeCo-SBIR: Semantically Consistent Prompt Learning for Zero-Shot Sketch-Based Image Retrieval

SeCo-SBIR：用于零样本草图-图像检索的语义一致提示学习

Long Hoang Dang, Tuan Nguyen Huu, Nguyen Minh Hieu, Tu Minh Phuong

AI总结 SeCo-SBIR是解决CLIP适配ZS-SBIR时域适配与泛化矛盾的语义一致提示学习框架，结合多模态提示、一致性约束与多目标损失，在三类ZS-SBIR基准上达最优性能。

Comments ACMMM 26

2608.03083 2026-08-05 cs.CV cs.CL 新提交

GSTEP: Global Spatio-Temporal Density-Driven Visual Token Pruning for Efficient Video Large Language Models

GSTEP：面向高效视频大语言模型的全局时空密度驱动视觉令牌剪枝

Mengjie Zhang, Qihui Zhu, Tao Zhang, Shuangwu Chen, Huihuang Qin, Yu Guo, Shenghao Ye, Zijian Wen, Yunpeng Hou, Dong Jin, Xiaobin Tan, Huasen He, Jian Yang

AI总结本文提出即插即用的GSTEP剪枝框架，解决现有视频令牌剪枝方法的局部剪枝缺陷，在多个VideoLLMs上实现良好的准确率-效率权衡，在LLaVA-OneVision-7B上剪去75%视觉令牌，保留100.2%原始性能并获1.17倍端到端加速。

Comments 4 figures, accepted to ACM MM 26'

2608.03047 2026-08-05 cs.CV cs.MM 新提交

AIDE: Automated Instruction via Distilled Expertise for Reference-Free Motor Skill Coaching

AIDE：基于提炼专业知识的自动化指令，用于无参考的运动技能指导

Yoshiki Ito

AI总结研究针对运动技能指导中专业教练稀缺的问题，提出AIDE框架，仅训练阶段用专业参考、推理阶段仅用学习者姿态生成反馈，在ExpertAF数据集上性能优于无参考基线，与需双阶段专业演示的方法相当。

Comments Accepted to ACM Multimedia 2026. 12 pages (including supplementary material), 4 figures