computer vision
PulseAugur coverage of computer vision — every cluster mentioning computer vision across labs, papers, and developer communities, ranked by signal.
8 天有情绪数据
-
New KAN variants tackle efficiency and hardware implementation
Researchers have developed a new variant of Kolmogorov-Arnold Networks (KANs) called Kolmogorov-Arnold Fourier Networks (KAFs) to address limitations in parameter efficiency and high-frequency feature capture. KAFs repa…
-
OpenCV 库支撑着现实世界的计算机视觉系统
OpenCV 是一个开源库，是现代人工智能模型出现之前就已存在并支持其运行的计算机视觉基础工具。它处理诸如清理和格式化视觉数据等基本预处理任务，这对于人工智能模型做出准确预测至关重要。该库广泛应用于从机器人到医学成像的各个行业，使系统能够解释和响应视觉信息。
-
大华将计算机视觉与 AI Agent 融合应用于工业领域
大华分享了其将计算机视觉（CV）与 AI Agent 融合以增强工业应用的策略，超越了简单的聊天机器人，发展为能够进行静默监控和自主决策的系统。该公司已开发了星汉大模型系列，包括用于视觉理解的 V 系列和用于行业特定逻辑的 L 系列，旨在弥合 AI 能力与行业理解之间的差距。大华的方法侧重于构建能够充当外部大脑的 AI Agent，在尊重数据安全的同时与现有数据和系统集成，最终使行业和个人受益于 AI。
-
合成数据在罕见病识别中可媲美真实世界表现
研究人员调查了仅使用合成数据通过面部表型识别儿科罕见病的有效性。他们的研究发现，当有足够的合成数据时，仅在合成图像上训练的模型取得了与仅使用真实数据训练的模型相当的性能。这表明高保真合成数据可以有效地近似真实世界分布，为医学教育和患者沟通提供一种保护隐私的资源。
-
Digi-Texx offers data annotation to boost AI development
Digi-Texx offers data annotation services to enhance AI development across various domains like computer vision and NLP. Their services aim to reduce algorithmic bias and improve the scalability of machine learning mode…
-
Computer Vision Services Target Miami Businesses
Codeponents offers computer vision solutions for Miami businesses, aiming to extract actionable insights from visual data. Their services are designed to automate analysis, reduce errors, and enhance operational efficie…
-
Yann LeCun 宣扬 JEPA 作为通用人工智能 (AGI) 的 LLM 替代方案
Yann LeCun 提出了联合嵌入预测架构 (JEPA) 作为实现通用人工智能 (AGI) 的大型语言模型 (LLMs) 的潜在替代方案。该方法旨在构建能够通过预测和表示学习来理解世界的 AI 系统，特别适用于机器人技术和计算机视觉应用。LeCun 认为，与当前的 LLM 范式相比，JEPA 可能为实现 AGI 提供一条更有效、更成功的路径。
-
新方法将图像分割与3D形状对应起来
研究人员开发了一种在图像分割和3D形状之间建立对应关系的新方法，解决了外观、几何形状和视点差异带来的挑战。该方法将2D模型中的深度视觉特征提炼到3D表面上，以计算图像像素和形状顶点之间的特征相似性。这使得在图像分割中识别出与3D形状特定区域对应的“最佳分割伙伴”，最终实现更准确、语义上更有意义的对齐。
-
New framework enhances physical adversarial attacks on vision models
Researchers have developed a new framework called JMOF to create more effective physical adversarial attacks against computer vision models. This framework addresses the issue of attacks overfitting to single models by …
-
MetaMoE 使用公共代理数据统一私有 MoE 模型
研究人员推出 MetaMoE，一个新颖的框架，旨在统一独立训练的专家混合（MoE）模型，而无需访问私有客户端数据。该系统利用公共代理数据来近似私有分布并指导路由器和专家的训练。这种具有多样性意识的代理选择方法旨在改善专家协调和选择，在计算机视觉和自然语言处理任务的实验中优于现有的隐私保护 MoE 统一技术。
-
商汤科技在上海部署人工智能机器人，打造无人便利店
商汤科技在上海推出了其“烧麦狗”具身智能便利店，店内配备的机器人可以在短短15秒内完成交易。这些人工智能驱动的便利店旨在管理所有零售任务，从接单、补货到数据分析和运营决策。该公司希望通过其在计算机视觉、大模型和机器人技术方面的专业知识，加速到2026年底在全国各地广泛推广这些无人便利店。
-
具身AI在CVPR 2026上重新定义计算机视觉的角色
具身AI正将计算机视觉研究的焦点从理解静态图像转移到使智能体能够与现实世界互动和操纵。这一在CVPR 2026上显现的范式转变，通过要求系统不仅要看，还要根据反馈进行行动和学习，重新定义了计算机视觉的价值。从证明存在到扩展能力，这一演变使视觉成为物理智能的基础设施，改变了该领域内问题的定义、评估和解决方式。
-
Computer vision framework quantifies fish communities and biomass
Researchers have developed a new computer vision framework to automatically quantify fish communities and their biomass from underwater video. This method uses deep learning for fish identification, tracking, and 3D rec…
-
Art history field embraces computational methods and AI
A call for papers has been issued for a special issue on "Computational Approaches to Art" within the Computational Humanities Research journal. This initiative signals a significant shift in the field of art history. T…
-
AI professionals urged to optimize skills section for job visibility
In the AI field, professionals often neglect their skills section on platforms like Mastodon, which functions as valuable free advertising space. Underutilizing this section by listing only a few items can lead to reduc…
-
Robotics researchers unify factor graphs and certifiable estimation for optimization
Researchers have developed a unified framework for certifiable factor graph optimization, synthesizing previously independent paradigms. This new approach leverages the structure of factor graphs to apply certifiable es…
-
AIDA-ReID paper introduces adaptive intermediate domain adaptation for person re-identification
Researchers have introduced AIDA-ReID, a novel framework for person re-identification that addresses domain shifts in unseen environments. This adaptive intermediate domain adaptation approach dynamically controls featu…
-
New 'sphere cloud' method enhances privacy in 3D visual localization
Researchers have developed a novel privacy-preserving technique for visual localization using a "sphere cloud" representation. This method addresses concerns about deep neural networks reconstructing private maps from 3…
-
Survey maps dynamic neural networks for computer vision and sensor fusion
This survey paper provides a comprehensive overview of Dynamic Neural Networks (DNNs), focusing on their application in computer vision and multi-modal sensor fusion. It addresses the challenge of deploying large models…
-
新的VCON框架可实现平滑迭代的深度神经网络压缩，准确率损失极小
研究人员推出了一种名为消失的贡献（VCON）的新型框架，旨在简化深度神经网络的压缩过程。VCON通过在微调期间并行运行原始模型和压缩模型，实现了向压缩模型的更平滑、迭代的过渡。这种方法逐渐减少了未压缩模型的影响，同时增加了压缩模型的贡献，从而提高了稳定性和降低了准确率损失。在计算机视觉和自然语言处理任务上的评估表明，VCON持续提高了性能，典型准确率提升超过1%，某些配置的提升超过15%。