国产替代系列：富煌钢构拟收购中科视界布局第二曲 线，看好高速视觉领域需求成长及国产替代加速 高速摄像机市场高速增长，国产替代需求进一步提升 中科视界是国内少数能够自主生产商品化高速摄像仪的企业，2017 年公司 发布了国产首台万帧级高速相机。高速视觉是一种基于图像处理和计算机视 觉技术的视觉感知技术，公司下游主要为科研客户、工业客户以及军工客户。 2022 年中国高速机器视觉行业市场规模约 100 亿元，其中高速摄像机整机 约占 32 亿元，约占 32%，在新兴技术发展、下游市场增长、国产化替代进 程加快等因素的驱动下，预计 2023-2028 年年均复合增速约为 22%，到 2028 年中国高速机器视觉行业市场规模或将突破 330 亿元。 行业报告 | 行业专题研究 建筑装饰 证券研究报告 高速视觉领域技术壁垒较高，中科视界市场份额快速提升 高速机器视觉行业具有较高的技术壁垒，头部主要为 Phantom、Photron 等 国际高速摄像机龙头，在美国对高速摄像仪进行出口管制背景下，公司产品 市占率由 2019 年 8.6%提升至 2022 年的 22.2%，公司已基本拥有和第一梯队 企业竞争的实力产品，我们 ...

MMLongBench团队 投稿 量子位 | 公众号 QbitAI 多模态长文本理解 有综合性的评判标准了！ 来自香港科技大学、腾讯西雅图AI Lab、爱丁堡大学、Miniml.AI、英伟达的研究者联合提出了 MMLongBench ，旨在全面评估多模态模型 的长文本理解能力。 随着多模态大模型的单次推理的文本窗口快速提升，长上下文视觉-语言模型（Long-Context Vision-Language Models; LCVLMs）应运而 生，使模型能够在单次推理中处理数百张图像与较长的交错文本。 但当前，由于评估多模态长文本的基准测试稀缺，现有的测试集仅关注单个任务，比如大海捞针或者长文档问答。目前尚不清楚现有的模型在 长上下文环境下的 综合表现 ，具体在哪些任务上存在短板，以及它们对不同输入长度变化的适应能力究竟如何。 结果显示，无论闭源还是开源模型，在长上下文视觉-语言任务上都面临较大挑战 ，仍有巨大的提升空间。 此外，进一步的错误分析表明，(1) OCR能力和 (2) 跨模态检索能力仍然是当前LCVLMs在处理长文本时的瓶颈。 多任务多模态长文本测试集 多任务的数据构建 MMLongBench是一个 ...

Core Insights - BMW Group is set to officially implement its long-developed Horizon panoramic display technology in new generation vehicles, transforming the entire windshield into a display interface that surpasses existing HUD experiences [1][2] - The technology features ultra-close projection, achieving a panoramic display effect from A-pillar to A-pillar, with a display area of 40 inches, supporting 4K quality output and a contrast ratio of 100,000:1 [1] - The core of this technology lies in a patented nano-black coating that effectively eliminates reflections and glare, ensuring display effectiveness in complex lighting environments [1] Technology Development - BMW's involvement in HUD technology began in 2003, making it the first European manufacturer to introduce heads-up display technology in mass-produced vehicles [1] - The company applied for a patent for the panoramic display technology in 2021 and launched the BMW Panoramic Vision Bridge concept in 2023, indicating a systematic advancement based on a long-term technological strategy [1] User Experience and Research - BMW's China R&D team initiated usability studies in 2020, with over 1,180 users participating and more than 4,700 hours of interview research conducted [1] - The BMW SkyLab Human-Computer Interaction User Experience Research Center completed three years of digital validation work, training algorithms based on anonymized data from 6 million Chinese users [1] Design and Market Implications - The technology employs a "visual cone" design concept, layering critical information such as driving, navigation, and alerts according to the natural line of sight, reducing the frequency and time of visual focus switching for drivers [2] - The introduction of panoramic display technology reflects intensified competition in the smart cockpit sector of the automotive industry, significantly expanding the information display space compared to traditional HUD technology [2] - BMW plans to launch the first new generation SAV model equipped with the panoramic iDrive system by 2025, considering technological maturity and market demand for smart cockpit innovations [2]

论文第一作者林宏彬来自香港中文大学(深圳)理工学院的Deep Bit 实验室、深圳市未来智联网络研究院，导师为李镇老师。目前实验室的研究方向包括：自动驾 驶、医学成像和分子理解的多模态数据分析和生成等。 论文标题： DriveGEN: Generalized and Robust 3D Detection in Driving via Controllable Text-to-Image Diffusion Generation 论文链接： https://www.arxiv.org/abs/2503.11122 GitHub： https://github.com/Hongbin98/DriveGEN 任务背景 随着新能源汽车产业的持续发展，智能驾驶辅助技术的应用越来越广泛。其中，基于纯视觉的自动驾驶方案只需使用多视角图像进行环境感知与分析，具有 成本低、效率高的优势，因而备受关注。然而在实际应用中，视觉感知模型的泛化能力至关重要。 来自香港中文大学（深圳）等单位的学者们提出了一种名为 DriveGEN 的无训练自动驾驶图像可控生成方法。该方法无需额外训练生成模型，即可实现训 练图像数据的可控扩充，从而以较 ...

编辑丨王多鱼 排版丨水成文 光 在向生物传递大量外部信息方面起着尤为关键的作用，使生物能够理解世界。然而，哺乳动物只能感知 电磁波谱中很小一部分作为可见光，通常在 400-700 纳米的范围内。这意味着超过一半的太阳辐射能量以 红外线 （>700 纳米） 的形式存在，对哺乳动物来说是不可察觉的。 人眼所见光谱范围的局限是由视网膜感光细胞中的感光蛋白 （Opsin） 固有的物理化学特性所决定，这导 致了大量本可能获取到的感觉信息的缺失。尽管诸如夜视镜或红外光-可见光转换器之类的工具已被用于红 外探测，但它们需要额外的能量支持，并且通常无法区分多个光谱中的红外光信息。此外，每个红外光-可 见光转换器都需要多层结构，这使得它们不透明且难以与人眼集成。 2019 年， 薛天 团队等在 Cell 发表论文 【1】 ， 利用一种转换红外光成为可见光的上转换纳米材料，经特 殊修饰后注射到小鼠视网膜中，首次实现了哺乳动物的裸眼近红外 （NIR） 图像视觉能力。然而，由于手 术具有侵入性，这种方式显然不会被人们轻易接受。 因此， 通过非侵入性方式相对自由的调节人眼感光波谱范围， 甚至赋予人类近红外视觉能力，对人类而言 仍然至关 ...

Core Viewpoint - The company, Aobo Zhongguang, has achieved significant growth in revenue and profitability, driven by its focus on robotics and AI visual technology, establishing itself as a leader in the industry [2][3]. Group 1: Financial Performance - In Q1 2025, Aobo Zhongguang reported revenue of 191 million yuan, a year-on-year increase of 105.63%, and a net profit of approximately 24.32 million yuan, marking a turnaround to profitability [2]. - The production volume of 3D visual sensors reached 888,400 units in 2024, reflecting a year-on-year growth of 17.1% [5]. Group 2: Technological Advancements - The company has developed and mass-produced five generations of deep engine chips and multiple types of depth-sensing chips, maintaining an annual iteration pace [2]. - Aobo Zhongguang possesses core technological capabilities in chip design, optics, and algorithms, with a comprehensive product development capability across various technical routes including structured light, iToF, dToF, and Lidar [3]. Group 3: Market Position and Expansion - Aobo Zhongguang has established a dominant position in the domestic service robot vision market and is actively expanding into industrial robotics and humanoid robots, leveraging its first-mover advantage [4]. - The company is constructing a "3D Visual Perception Industry Intelligent Manufacturing Base" in Shunde, with a building area exceeding 100,000 square meters, to enhance its supply chain and production capacity [5]. Group 4: Strategic Collaborations - The company has formed stable ecological partnerships with international giants such as Microsoft, NVIDIA, and AMD, integrating its products into various developer ecosystems [2]. - Notably, the Femto Bolt depth camera from Aobo Zhongguang was utilized in a breakthrough project in the field of spatial intelligence by Stanford University, showcasing its technology's application in optimizing robotic actions [3].

Core Viewpoint - The company plans to increase its investment in its wholly-owned subsidiary, Halo Microelectronics (Hong Kong) Co., Limited, by USD 30 million to support the development of its smart visual perception business and meet the operational needs of its overseas subsidiaries [1][2]. Group 1: Investment Overview - The investment amount is USD 30 million, which is approximately RMB 215.97 million based on real-time exchange rates [2]. - After the investment, the total investment in Halo Microelectronics will increase from USD 90,001,300 to USD 120,001,300, maintaining a 100% ownership stake [2]. - The decision was approved by the company's board of directors on May 22, 2025, and does not require shareholder approval [2]. Group 2: Subsidiary Information - Halo Microelectronics (Hong Kong) Co., Limited is a limited liability company established on October 4, 2013, and is located in Hong Kong [3][4]. - The company specializes in logistics, procurement, and sales of integrated circuit analog chips [4]. - As of March 31, 2025, the total assets of Halo Microelectronics were RMB 96,535.87 million, with total liabilities of RMB 34,846.93 million and net assets of RMB 61,688.95 million [4]. Group 3: Financial Performance - For the fiscal year 2024, Halo Microelectronics reported a revenue of RMB 53,654.19 million and a net loss of RMB 9,449.41 million [4]. - In the first quarter of 2025, the company generated revenue of RMB 17,924.23 million with a net loss of RMB 998.46 million [4]. Group 4: Impact of Investment - The investment aims to enhance the capital strength of Halo Microelectronics and improve the operational capabilities and sustainable development of the company and its subsidiaries [5]. - The investment will not change the consolidation scope of the financial statements, as Halo Microelectronics remains a wholly-owned subsidiary [5]. - The investment is expected to have no significant adverse impact on the company's financial and operational status [5].

继本周"誓死捍卫"特斯拉之后，马斯克的又一句话透露出公司未来发展的关键信息。 近期，马斯克在接受CNBC访谈时明确表示： "从长期来看，唯一重要的是自动驾驶和Optimus。" 然而，摩根士丹利最新发布的报告却揭示了一个可能改变游戏规则的新对手——小米YU7。 据追风交易台消息，摩根士丹利21日发布研报表示，如果观察小米YU7的设计和规格，将对特斯拉非常有启发性。小米YU7外观像法拉利或阿斯顿马丁 SUV，价格却与丰田凯美瑞相当。 YU7的出现不仅为消费者提供了更多的选择，也给特斯拉带来了前所未有的压力。这引发一个关键问题：特斯拉是否应该继续推出更多传统方向盘式电动 车，还是应该专注于自动驾驶和机器人等前沿技术领域？ 40天内推出无监督Robotaxi！奥斯汀先行，扩张在即 摩根士丹利指出，马斯克在采访中确认，将在40天内在德克萨斯州奥斯汀推出完全自动驾驶的Cybercab，称"我们有数千辆正在测试的汽车"。 虽然初期部署规模很小（第一周仅10辆车），但马斯克预计"几个月内可能达到1000辆"，随后将扩展到旧金山、洛杉矶和圣安东尼奥等城市。 这符合摩根士丹利的预期，即特斯拉将优先在德克萨斯、内华达、加利福尼亚 ...

Core Insights - The article discusses the advancements and competitive landscape of the assisted driving industry, highlighting various companies' self-developed systems and strategies [1][4]. Group 1: Company Developments - Li Auto has launched its new generation dual-system intelligent driving solution, focusing on upgrading driving capabilities and synchronizing updates for smart electric vehicles [3]. - NIO's intelligent assisted driving system has reportedly avoided over 3.5 million collision risks, accumulating a total driving mileage of approximately 4.94 billion kilometers as of May 15, 2025 [3]. - Chery's Hawk 500 has achieved widespread adoption of assisted driving features, with the Hawk 700 targeting mid-to-high-end models and the Hawk 900 positioned as a flagship [3]. - GAC Group's GSD intelligent driving assistance system has accumulated 5 million user driving scenarios and over 40 million kilometers of high-level autonomous driving data [3]. Group 2: Industry Trends - BYD and XPeng are recognized as leaders in self-developed intelligent driving systems, with BYD's high-end system named "Tianshen Eye" [4]. - Bosch's China president has expressed skepticism about the self-development model, suggesting that mid-level intelligent driving should become standard and that costs could be better managed through supply chain partnerships [4]. - Huawei is positioned as a top player in the intelligent driving system market, with plans for 10 brands from 7 automakers to adopt its solutions, potentially exceeding 500,000 vehicles [4][5]. - Huawei's collaboration models include component supply, Huawei Inside (HI) partnerships, and deep cooperation with automakers, with the latter being the most integrated approach [5]. Group 3: Strategic Partnerships - SAIC Group has publicly stated its intention to maintain control over core technologies while also choosing to collaborate with Huawei [6]. - The partnerships with Huawei have led to increased sales for collaborating automakers, but questions remain about their ability to independently develop high-quality vehicles [6].

Group 1: Core Perspectives - Tesla emphasizes the importance of its vision processing solution, stating that it aims to make safe and intelligent products affordable for everyone [1] - Tesla's upcoming Full Self-Driving (FSD) solution will rely solely on artificial intelligence and a vision-first strategy, abandoning LiDAR technology [1][4] - The global market for automotive LiDAR is projected to grow significantly, with a 68% increase expected in 2024, reaching a market size of $692 million [1] Group 2: Technology and Market Dynamics - The debate between pure vision systems and multi-sensor fusion approaches continues, reflecting a complex interplay of technology, cost logic, and market strategies [2] - Tesla's vision processing system, trained on billions of real-world data samples, aims to achieve safer driving through a neural network architecture [4] - The pure vision approach is characterized by its reliance on cameras, which reduces system integration complexity and hardware costs, but faces challenges in adverse weather conditions [6] Group 3: Industry Comparisons - In China, many automakers are developing intelligent driving technologies tailored to local road conditions, which may outperform Tesla's pure vision approach [7] - The safety redundancy provided by LiDAR is highlighted, especially in complex driving scenarios where visual systems may fail [16] - The divergence in strategies between Tesla and Chinese automakers represents a fundamental debate between algorithm-driven and hardware-driven approaches [18] Group 4: Sensor Technology - The advantages and disadvantages of various sensors, including cameras, ultrasonic, millimeter-wave, and LiDAR, are outlined, emphasizing the need for multi-sensor integration for enhanced safety [11][12][13] - LiDAR's high precision and ability to operate in various lighting conditions make it suitable for complex urban environments [12] - The integration of multiple sensors can enhance the robustness of intelligent driving systems, addressing the limitations of single-sensor approaches [17] Group 5: Future Trends - The cost of LiDAR technology has decreased significantly, making it more accessible for a wider range of vehicles, thus driving the adoption of advanced driver-assistance systems [19] - The industry is moving towards a more interconnected system of intelligent driving, leveraging AI networks and real-time data sharing for improved decision-making [19] - Safety remains a paramount concern in the development of intelligent driving technologies, with a focus on building reliable systems that users can trust [20]