图片来源： Buildots 在建筑行业中，管理人员很容易与现场实际情况脱节。他们需要同时处理多项任务，包括掌握成本动 态、与所有利益相关方沟通，以及评估与承包商账单和绩效等方面相关的风险。 Buildots 希望通过人工智能和计算机视觉技术彻底改变这一现状。 这家芝加哥初创公司由 Roy Danon 、 Aviv Leibovici 和 Yakir Sudry 于 2018 年联合创立，其平台通 过处理管理人员安全帽上 360 度摄像头拍摄的图像来追踪施工进度。 Buildots 在由 Qumra Capital 领投的 D 轮融资中筹集了 4500 万美元，OG Venture Partners、TLV Partners、Poalim Equity、Future Energy Ventures 和 Viola Growth 跟投。此次融资使该公司总融资额 达到 1.66 亿美元。 该系统不仅具备监测功能，还能进行预测。团队可以通过 AI 聊天机器人查询项目状态，并使用预测 工具获取可能延误风险或进度问题的预警，避免这些问题演变成代价高昂的麻烦。 我们正在招募新一期的实习生 我们正在寻找有创造力的00后创业 ...

Group 1 - The world's first humanoid robot fighting competition was held in Hangzhou, attracting significant attention from industry professionals [1] - Machine vision is a key technology for humanoid robots to perceive and understand their environment, with domestic manufacturers primarily using 3D vision combined with LiDAR to enhance environmental modeling and navigation accuracy [1] - Orbbec, a leading company in the domestic service robot vision market, holds over 70% market share and is one of the earliest players in the robot vision sector [1] Group 2 - Current advancements in humanoid robots focus on joint flexibility and balance control, with the execution of commands exceeding expectations during the competition [2] - For true autonomous robot fighting, the next step involves upgrading from passive execution to active execution, where visual sensor data becomes crucial for decision-making [2] - Orbbec is one of the few companies globally that has comprehensively developed six major 3D visual perception technology routes, actively providing samples of its binocular structured light products for compatibility with most humanoid robot clients [2]

Core Insights - The recent G1 robot fighting competition has sparked widespread discussion in the industry, showcasing advancements in robot technology and exceeding market expectations regarding robot stability and impact resistance, potentially leading to a new wave in the robotics sector [1][4] Group 1: Technology and Innovation - The implementation of robot fighting relies on various sensing technologies, including force sensors, tactile sensors, and visual solutions, with visual perception technology being crucial for robots to understand their environment [1][2] - The G1 robots are equipped with dual-depth cameras and 3D LiDAR, enabling 360-degree environmental perception and real-time posture adjustments through multi-sensor fusion technology [2][3] - The current competition still utilizes manual remote control, indicating that the performance of robot teams depends on the collaboration between human operators and robotic participants [3] Group 2: Market Position and Competitive Landscape - Aobi Zhongguang, a leading player in the domestic robot vision market, holds over 70% market share and is one of the earliest companies to focus on robot vision technology [2][4] - The upcoming "Mecha King" competition in December will be the first to feature humanoid robots as the main competitors, promoting the integration of robotics with sports and culture [4][5] - The rise of robot fighting competitions is expected to accelerate technological advancements and commercial applications in the humanoid robot industry, with domestic 3D visual perception technology and multi-dimensional sensing technology playing significant roles [4][5]

机器之心报道 机器之心编辑部 当今计算机视觉领域最热门的三个方向。 当今计算机视觉领域最热门的话题有哪些？ 「自 2020 年 NeRF 论文首次发表以来，利用深度网络进行 3D 重建已成为趋势。如今高斯泼溅（Gaussian splatting）技术进一步推动了这一发展，」CVPR 2025 程序联合主席、美国俄勒冈州立大学副教授 Fuxin Li 分享道，「本质上，计算机视觉与图形学正在融合。神经渲染研究的兴起，显著推动了 3D 相关论文数量的 增长。」 图像与视频合成 随着研究的不断发展，学界现在能够通过视频和图像形式更精确地呈现环境信息。对该领域的探索已成为 CVPR 2025 论文的焦点，图像和视频合成成为今年大会上最大的类别之一。 「今年商业聊天机器人的一大趋势是它们已经实现了多模态化；它们现在不仅能分析和生成文本，还能分 析和生成图像，有时还能生成视频，」Isola 解释道。「即将出现的能力是生成完整的交互式世界。CVPR 刚刚，CVPR 官方给出了最新统计，他们根据论文提交情况，总结出三大方向： 这份统计是基于全球 4 万多名作者 13008 份投稿结果产生的。相比往年，今年的投稿数量增长了 ...

Core Insights - In January 2025, a significant influx of overseas users began to engage with Xiaohongshu, leading to a unique cultural exchange where users shared pet experiences, assisted with English homework, and learned Chinese cooking from Chinese users [1][2] - The primary challenge faced by Xiaohongshu was the language barrier, prompting the need for effective communication tools to facilitate user interactions [1] Group 1: Technological Developments - Xiaohongshu quickly developed a "one-click translation" feature in response to user demands, allowing automatic translation of English comments into Chinese, thus streamlining the user experience [1] - The translation functionality is built on a multi-modal AI model that integrates Natural Language Processing (NLP), Optical Character Recognition (OCR), and Computer Vision (CV), enabling the system to understand not just text but also cultural nuances such as memes [1] - A dynamic learning mechanism is in place where user edits to translations contribute to ongoing model training, particularly enhancing the understanding of culturally sensitive content [1] Group 2: Cultural Integration - The company emphasizes that its translation capabilities extend beyond mere word-for-word translation to encompass cultural adaptation, reflecting the diversity of human civilization [1] - Xiaohongshu's approach illustrates the importance of embedding technology within a humanistic framework, transforming barriers into bridges for communication [2]

Core Viewpoint - Jiangsu Zejing Automotive Electronics Co., Ltd. is pursuing an IPO in Hong Kong, facing challenges such as losses, high asset-liability ratio, and pressure from downstream clients for price reductions [1][5]. Group 1: Financial Performance - The main product, HUD solutions, has seen a continuous decline in average price from 974.31 RMB in 2022 to 865.47 RMB in 2024, while the gross margin has improved from 22.6% in 2022 to 27.3% in 2024 [2][3]. - The sales volume of HUD solutions increased significantly from approximately 17.57 thousand units in 2022 to 62.46 thousand units in 2024, representing a growth of 2.55 times [2][3]. - Despite the increase in sales volume, the company's labor costs decreased by 12.82% in 2024, amounting to 25.33 million RMB [3]. Group 2: Market Position - Zejing Electronics holds a market share of 16.2% in the Chinese HUD solutions market, ranking second among suppliers as of 2024 [3]. - The revenue growth rate slowed significantly from 156.6% in 2023 to 5.1% in 2024, indicating potential market challenges [3][5]. Group 3: Client Dependency and Revenue Sources - The top five clients contributed a substantial portion of the company's revenue, accounting for 93.0% in 2022, 93.8% in 2023, and 80.9% in 2024 [4]. - Revenue from the top five clients decreased by 9.13% in 2024, amounting to 468 million RMB compared to 515 million RMB in 2023 [5]. Group 4: Challenges Ahead - The company has not achieved profitability during the reporting period, with an asset-liability ratio of 215.6% and a current ratio of only 41.2% as of 2024 [6]. - Trade receivables reached 296 million RMB by the end of 2024, with an increase in the average collection period from 102.2 days in 2022 to 140.5 days in 2024 [6]. - The company may face ongoing pricing pressure from automotive manufacturers, which could adversely affect its business and financial performance [7].

本文由北京智源研究院多模态大模型研究中心（团队负责人王鑫龙，团队代表作 EMU 系列、EVA 系列、Painter & SegGPT）、中科院自动化所和大连理 工大学联合完成。 在多模态学习蓬勃发展的当下，视觉 tokenizer 作为连接视觉信息与下游任务的关键桥梁，其性能优劣直接决定了多模态模型的表现。然而，传统的视觉 tokenization 方法存在一个致命缺陷：视觉 tokenizer 的优化与下游任务的训练是相互割裂的。 这种分离式的训练范式假设视觉 tokens 能够在不同任务间无缝通用，但现实情况是，为低级重建任务优化的视觉 tokenizer 往往难以满足诸如图像生成、 视觉问答等需要丰富语义表示的下游任务需求，导致下游任务的性能受限。 针对这一亟待解决的问题，我们提出了 ETT（End-to-End Vision Tokenizer Tuning），一种全新的端到端视觉 tokenizer 调优方法。 ETT 的核心架构与训练策略 ETT 创新性地实现了视觉 tokenization 与目标自回归任务的联合优化，打破了传统方法中视觉 tokenizer 一旦训练完成便固定的常规，充分释放了 ...

Core Viewpoint - The evolution of security systems in the financial industry has transitioned from traditional methods to advanced intelligent video surveillance technologies, driven by developments in artificial intelligence and big data [3][4][19]. Group 1: Historical Development of Security Systems - The security systems in the financial sector date back hundreds of years, evolving with technological advancements and changing security needs [3]. - In the early 20th century, banks began implementing intrusion alarms and video monitoring systems to protect cash and valuables [3]. - By the early 21st century, video surveillance technology matured, becoming a crucial component of security systems, primarily for crime deterrence and evidence collection [3][4]. Group 2: Technological Advancements - The rapid development of computer vision technology in the early 2000s introduced capabilities such as target recognition and behavior analysis, enhancing video surveillance systems [4]. - The integration of artificial intelligence and machine learning has led to more accurate and efficient functionalities in video monitoring, including facial recognition and behavior analysis [4][5]. - Intelligent Video Analytics (IVA) automates the analysis of video data, allowing for proactive alerts and enhanced security measures [5][6]. Group 3: Key Technologies in Intelligent Video Surveillance - Video image processing technologies have evolved from traditional methods to advanced algorithms, enabling real-time monitoring and anomaly detection [6][7]. - Facial recognition technology has significantly advanced, utilizing deep learning to improve accuracy and expand application scenarios [8][9]. - Video structural analysis allows for the organization and retrieval of video content, transforming unstructured data into structured information for better analysis [10][11]. Group 4: Data Analysis and Integration - Multi-dimensional data analysis models combine video data with other business data, facilitating comprehensive assessments and predictive modeling in security applications [12]. - The classification of intelligent video analysis technologies can be based on algorithm types, including video improvement, analysis, recognition, and retrieval [13][14]. - The physical location of intelligent video products can be categorized into edge intelligence and cloud intelligence, each serving different analytical needs [17][18]. Group 5: Future Directions and Industry Implications - The financial industry must prioritize top-level design and planning for intelligent video systems, focusing on a combination of edge and cloud intelligence [19][23]. - The emergence of large-scale AI models is expected to drive significant advancements in video processing and understanding, although challenges remain in achieving high accuracy for security applications [24][25]. - A strategic approach to technology adoption is essential, balancing innovation with the potential risks associated with new technologies [25].

公司研究 证 券 研 究 报 告 思看科技（688583）深度研究报告 强推（首次） 从平面到立体，商业级蓝海启航，工业级技 术筑基 [ReportFinancialIndex] 主要财务指标 | | 2024A | 2025E | 2026E | 2027E | | --- | --- | --- | --- | --- | | 营业总收入(百万) | 333 | 421 | 534 | 677 | | 同比增速(%) | 22.4% | 26.5% | 26.9% | 26.8% | | 归母净利润(百万) | 121 | 144 | 173 | 209 | | 同比增速(%) | 5.5% | 19.6% | 19.8% | 20.8% | | 每股盈利(元) | 1.77 | 2.12 | 2.54 | 3.07 | | 市盈率(倍) | 60 | 50 | 42 | 35 | | 市净率(倍) | 11.5 | 9.6 | 5.3 | 4.8 | 资料来源：公司公告，华创证券预测 注：股价为 2025 年 5 月 23 日收盘价 仪器仪表Ⅲ 2025 年 05 月 24 日 | 目标价：148.35 ...

Core Viewpoint - The article highlights the emergence of innovative technology companies in Hangzhou, particularly focusing on the "Six Little Dragons" that are reshaping the private economy in the city through advancements in artificial intelligence and robotics. Group 1: Company Developments - Zhejiang Qiangnao Technology Co., Ltd. has achieved the global first mass production of 100,000 portable high-precision brain-computer interface devices, overcoming engineering and technical challenges in consumer-grade brain-computer interface equipment [2][5]. - Cloud Deep Technology Co., Ltd. has developed the "Jueying" X30 robotic dog, which has been deployed in various practical applications, including disaster relief and security inspections, achieving 95% domestic self-control [7][8]. - Hangzhou Blue Core Technology Co., Ltd. focuses on 3D visual perception technology for mobile robots, successfully integrating this technology into their products, which has led to significant improvements in operational efficiency in manufacturing environments [12][15]. Group 2: Government Support and Policies - The Hangzhou government has actively supported the development of brain-computer interface technology by providing research and industrialization space, as well as establishing laboratories for technical support [6][5]. - The "Spring Sunshine Plan" launched in 2024 aims to reduce costs for enterprises, particularly small and medium-sized enterprises, by providing financial support and streamlining policy implementation [18][19]. - Hangzhou's commitment to increasing annual fiscal investment in technology by over 15% demonstrates the city's determination to foster innovation and support the tech industry [17]. Group 3: Talent Attraction and Ecosystem - The talent policies in Hangzhou have attracted skilled professionals from top universities, contributing to the workforce of local tech companies, with many employees benefiting from housing subsidies and other incentives [21][22]. - The collaboration between local universities, such as Zhejiang University, and tech companies has created a robust ecosystem that supports innovation and entrepreneurship [10][27]. - The "entrepreneurial effect" generated by successful companies like Alibaba has inspired a wave of new startups in Hangzhou, showcasing the city's potential for tech-driven growth [25][26].