arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

Conference on Computer Vision and Pattern Recognition · 会议 · Computer Vision

2026-09-01 至 2026-09-01 共收录 8
2608.30653 2026-09-01 cs.CV cs.AI cs.LG 新提交

Fine-Grained Multi Image Object Hallucination Benchmark

细粒度多图像物体幻觉基准测试

Joonki Min, Chaeyun Kim, Hyungwook Choi, Yejin Kim, Kihyun Kim, Yohan Jo, Joonseok Lee

机构 * Seoul National University(首尔大学) AIM Intelligence(AIM智能公司)

AI总结 本研究针对多模态大语言模型的物体幻觉问题,推出细粒度多图像物体幻觉基准测试MIOH,评估29个模型后发现顶尖模型仍存在明显失败模式,为开发可靠多模态AI提供关键评估工具。

Comments Accepted at CVPR 2026

Journal ref Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026, pp. 18295-18305

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.29759 2026-09-01 cs.CV cs.AI 新提交

SynCrash: A Multi-Stage Pipeline for Zero-Shot Accident Detection and Localization in Traffic Surveillance Video

SynCrash:面向交通监控视频零样本事故检测与定位的多阶段流水线

Arkya Jyoti Bagchi, Ritul Jangir, Varun Raskar

机构 * Indian Institute of Technology Jodhpur(印度焦特布尔印度理工学院)

AI总结 提出SynCrash多阶段流水线,针对CVPR2026挑战赛ACCIDENT任务,在无真实标注数据下,结合VideoMAEv2、YOLO及物理启发式算法,实现交通监控视频零样本事故检测、定位与碰撞分类。

Comments Accepted at the CVPR 2026 AUTOPILOT Workshop (non-archival)

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.29252 2026-09-01 cs.AI 新提交

Dynamic Important Example Mining for Reinforcement Finetuning

用于强化微调的动态重要示例挖掘

Haoru Tan, Sitong Wu, Yanfeng Chen, Shizhen Zhao, Yang-Tian Sun, Tianjia Liu, Chirui Chang, Shaofeng Zhang, Samm Sun, Xiuzhe Wu, Ruobing Xie, Xiaojuan Qi

机构 * HKU(香港大学) Tencent(腾讯) CUHK(香港中文大学) Stanford(斯坦福大学)

AI总结 该研究针对强化微调中样本价值固定假设的缺陷,提出DIEM框架,整合梯度对齐重要性估计器与约束批量重加权方案,在多个推理基准中优于相关基线。

Journal ref CVPR-2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.07288 2026-09-01 cs.CV cs.GR 版本更新

ExMesh: Explicit Mesh Reconstruction with Topology Adaptation

ExMesh: 具有拓扑自适应的显式网格重建

Chuanjin Fan, Lifan Wu, Wenjie Chang, Hanzhi Chang, Wenfei Yang, Tianzhu Zhang

机构 * University of Science and Technology of China(中国科学技术大学) National Key Laboratory of Deep Space Exploration, Deep Space Exploration Laboratory(国家空间科学探测重点实验室,深空探测实验室)

AI总结 提出ExMesh框架,通过可微优化与离散拓扑更新直接优化显式网格,引入自适应顶点分裂合并和实时UV维护,实现从粗到细的优化,兼顾精度、效率和网格简洁性。

Comments Accepted at the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2026 (CVPR 2026); Project page: https://fan-treasure.github.io/ExMesh_page.github.io/

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.22455 2026-09-01 cs.CV cs.AI cs.LG physics.optics 版本更新

Making the Discrete Continuous: Synthetic RAW Augmentations for Fine-Grained Evaluation of Person Detection Performance in Low Light

使离散的成为连续的:合成RAW增强用于细粒度评估人检测性能在低光环境

Valeria Pais, Malena Mendilaharzu, Daniele Faccio, Luis Oala, Christoph Clausen, Bruno Sanguinetti

机构 * University of Glasgow(格拉斯哥大学) Dotphoton

AI总结 本文提出了一种合成RAW增强方法,用于在低光条件下更准确地评估人检测模型的性能,通过生成与相机传感器噪声模型匹配的低光样本,以改善基准测试的数据覆盖。

Comments Accepted non-archival paper at the CVPR 2026 AUTOPILOT Workshop (Autonomous Understanding Through Open-world Perception and Integrated Language Models for On-road Tasks)

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.11521 2026-09-01 cs.CV cs.AI 版本更新

EReCu: Pseudo-label Evolution Fusion and Refinement with Multi-Cue Learning for Unsupervised Camouflage Detection

EReCu: 多线索学习下的伪标签演化融合与精炼用于无监督伪装检测

Shuo Jiang, Gaojia Zhang, Min Tan, Yufei Yin, Gang Pan

AI总结 EReCu提出了一种统一的无监督伪装检测框架,通过多线索学习提升伪标签可靠性与特征保真度,结合原生感知模块、伪标签演化融合和局部精炼,实现高精度的伪装检测。

Comments Accepted by CVPR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.04825 2026-09-01 cs.CV cs.LG

Mitigating Instance Entanglement in Instance-Dependent Partial Label Learning

缓解实例依赖部分标签学习中的实例纠缠

Rui Zhao, Bin Shi, Kai Sun, Bo Dong

机构 * School of Computer Science and Technology, Xi’an Jiaotong University(西安交通大学计算机科学与技术学院) Shaanxi Province Key Laboratory of Big Data Knowledge Engineering, Xi’an Jiaotong University(陕西省大数据知识工程重点实验室) School of Distance Education, Xi’an Jiaotong University(西安交通大学继续教育学院)

AI总结 本文提出CAD框架,通过内在和类间调节缓解实例依赖部分标签学习中的实例纠缠问题,提升分类性能。

Comments Accepted to CVPR2026

Journal ref Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026, pp. 39659-39668

详情

展开后加载摘要…

URL PDF HTML 收藏
2412.11061 2026-09-01 cs.CV cs.CY cs.LG 版本更新

Classification Drives Geographic Bias in Street Scene Segmentation

分类导致街景分割中的地理偏差

Rahul Nair, Gabriel Tseng, Esther Rolf, Bhanu Tokas, Hannah Kerner

机构 * Arizona State University(亚利桑那州立大学) Mila Quebec AI Institute(米拉魁北克人工智能研究所) University of Colorado Boulder(科罗拉多大学博尔德分校)

AI总结 该研究针对实例分割任务,发现以欧洲驾驶场景训练的模型存在地理偏差,且偏差源于分类错误,通过粗粒度类别分组可缓解此偏差。

Comments Accepted at the CVPR 2025 Workshop on Fair, Data-Efficient, and Trusted Computer Vision

Journal ref R. Nair, B. Tokas, G. Tseng, E. Rolf and H. Kerner, "Classification Drives Geographic Bias in Street Scene Segmentation," 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

详情

展开后加载摘要…

URL PDF HTML 收藏