Shiliang Pu

Organized by Hikvision, the 2018 AI Cloud World Summit was successfully held at Hangzhou International Expo Centre from March 30–31. Over 2,000 participants attended the summit, including industry experts and scholars, as well as Hikvision partners including Microsoft, Intel, Milestone Systems and Axxonsoft. Attendees and presenters explored trends in AI development, its real-world applications, and opportunities for AI-related partner cooperation in global markets. Hikvision introduced AI Cloud in October 2017 at the China Public Security Expo in Shenzhen. AI Cloud is a development concept born out of the Internet-of-Things (IoT) era. It is a three-layered architecture that incorporates cloud and edge computing to provide multi-dimensional perception and front-end processing at Edge Nodes, and then process data in real-time and converged to Edge Domains for intelligent applications and create new data, and further converge on-demand data to the Cloud Center for big data analysis. The next generation of Artificial Intelligence is moving towards big data intelligence, collective intelligence, and cross-media intelligence AI cloud Ecosystems Through Open Collaboration Artificial intelligence is thought to be another rising technology that is going to significantly impact the human society. In his keynote speech, Yunhe Pan, a member of the Chinese Academy of Engineering, highlighted that Artificial Intelligence is already evolving to its next stage due to the emergence of new social needs such as smart cities, smart medicine, and unmanned driving, as well as significant changes in information environments like the Internet, mobile computing, cloud computing, Internet of Things and others. In short, the next generation of Artificial Intelligence is moving towards big data intelligence, collective intelligence, and cross-media intelligence. “A large number of application requirements are surging in a fragmented way. The implementation of AI is restricted mainly by three factors – data, computing power and application. To accelerate the process, it is necessary for all stakeholders in every industry to be open and collaborative to build an AI industry ecosystem,” said Yangzhong Hu, President of Hikvision addressing in the summit. “To enable more partners to participate in the construction and sharing of the AI ecosystem, the company will be fully open on its products and services, including the AI Cloud software platform, training system, AI services, data annotation and data sharing services, etc.” Dr. Shiliang Pu pointed to advantages of edge perception and its ability to generate multi-layered cognition to empower AI applications Emerging Needs For Edge Computing At the summit, Oliver Philippou, Associate Research Director at IHS Markit, gave Hikvision’s international attendees a presentation titled ‘Artificial Intelligence – The Present and Future Prospects for Video Surveillance’. According to Philippou, increasing numbers of network cameras on the global market pushes the development of video analytics further toward convergence – to edge and cloud systems – requiring better algorithms and deep learning to provide maximum accuracy. Dr. Shiliang Pu, President of Hikvision’s Research Institute, shared the company’s AI research facilities, milestones, and achievements over the past years. Dr. Pu pointed to advantages of edge perception and its ability to generate multi-layered cognition to empower AI applications. It is the joint efforts for various providers to build up this AI ecosystem. Hikvision was honoured to have three of its strategic partners – Intel, Milestone Systems, and Axxonsoft – give keynote speeches in a summit session about mutual partnership and how AI changes partner ecosystems.

Paving the way for the creation of AI cities, NVIDIA has unveiled the NVIDIA Metropolis™ intelligent video analytics platform. Metropolis Deep Learning Metropolis makes cities safer and smarter by applying deep learning to video streams for applications such as public safety, traffic management and resource optimization. More than 50 NVIDIA AI city partner companies are already providing products and applications that use deep learning on GPUs, many of which will be on display this week at the GPU Technology Conference. “Deep learning is enabling powerful intelligent video analytics that turn anonymized video into real-time valuable insights, enhancing safety and improving lives,” said Deepu Talla, vice president and general manager of the Tegra business at NVIDIA. “The NVIDIA Metropolis platform enables customers to put AI behind every video stream to create smarter cities.” Intelligent Video Analytics Video is the world’s largest generator of data, captured by hundreds of millions of cameras deployed in areas such as government property, public transit, commercial buildings and roadways. By 2020, the cumulative number of cameras is expected to rise to approximately 1 billion. Humans currently monitor only a fraction of captured video, with most stored on disks for later review. Initial efforts at real-time video analytics techniques have proved far less reliable than human interpretation. Intelligent video analytics solves this challenge by using deep learning in cameras, on-premises video recorders and servers, and in the cloud to monitor video instantaneously with accuracy and scalability. Metropolis spans multiple NVIDIA products that operate on a unified architecture. High performance deep learning inferencing happens at the edge with the NVIDIA Jetson™ embedded computing platform, and through servers and data centers with NVIDIA® Tesla® GPU accelerators. Rich data visualization is powered by NVIDIA Quadro® professional graphics and the entire edge-to-cloud platform is supported by NVIDIA’s rich software development kits, including JetPack, DeepStream and TensorRT™. “NVIDIA’s end-to-end Metropolis platform can be applied to video streams to create smarter and safer applications for a variety of industries" Growing AI City Partner Support More than 50 NVIDIA AI city partners already help customers reveal insights and take real-time action using deep learning on NVIDIA GPUs. Among them are industry leaders such as Avigilon, Dahua, Hanwha Techwin, Hikvision and Milestone. “With the fast-paced environment of a city, there are a near infinite number of activities taking place,” said Dr. Mahesh Saptharishi, chief technology officer at Avigilon. “We’re excited by the potential of NVIDIA’s Metropolis platform, as Avigilon continues to deliver AI-powered surveillance solutions and video analytics that focus users’ attention on what matters most, in order to take action.” “NVIDIA’s end-to-end Metropolis platform can be applied to video streams to create smarter and safer applications for a variety of industries – from transportation to commercial,” said Shiliang Pu, president at Hikvision Research Institute. “The benefit of GPU deep learning is that data can be analyzed quickly and accurately to drive deeper insights.” “City management customers using Milestone’s upcoming Video Processing Server with NVIDIA Metropolis are positioned to take the lead in the adoption of deep learning for video-enabled IoT devices,” said Bjørn Skou Eilertsen, chief technology officer at Milestone Systems. “Unleashing the value of this metadata will provide intelligent insights to take smart action.”

The challenge attracts participants from more than fifty leading research institutions worldwide Hikvision, a supplier of innovative video surveillance products and solutions, recently achieved the no.1 position in the Scene classification category at the ImageNet Large Scale Visual Recognition Challenge 2016. Object Category Classification And Detection ImageNet, the global image database resource, hosts the Large Scale Visual Recognition Challenge (ILSVRC) to establish a benchmark in object category classification and detection across hundreds of object categories, and millions of images. Run annually since 2010, the challenge attracts participants from more than fifty leading research institutions worldwide. Organized by Stanford University, Carnegie Mellon University, University of Michigan, and UNC Chapel Hill, a high level impetus for the ILSVRC is to allow researchers to compare progress in detection across a wider variety of objects. Another driver is to measure the progress of computer vision for large-scale image indexing, to enable effective retrieval and annotation. No.1 In Scene Classification Category Encompassing Object detection, Object localization, Object detection from video, Scene classification, and Scene parsing, the five ILSVRC categories were each considered and evaluated against a precise challenge. For the Scene classification task, where Hikvision achieved no.1 position, Hikvision Research Institute used inception-style networks and not-so-deep residuals networks that perform better in considerably less training time, according to Hikvision’s experiments and several improvements made for training and testing. “Since it was established, Hikvision Research Institute has accumulated a deep technical background” “The technical data resulting from the competition can be applied to vehicle detection, license plate recognition, vehicle sub-brand recognition, human detection, human property analysis, face recognition, image search and much more, to greatly enhance product performance and application results,” enthused Shiliang Pu, Executive Vice President at Hikvision Research Institute. “In the future, the intelligence and automation levels of machines will be improved significantly, and utilized in the sectors of intelligent surveillance, driver assistance system, intelligent traffic sensing, robotics and unmanned aerial vehicles, to name but a few.” Hikvision Research Institute “Since it was established, Hikvision Research Institute has accumulated a deep technical background,” comments Cynthia Ho, Vice President of Hikvision. “The results of the LSVRC underline Hikvision Research Institute is at the forefront of computer vision research, and their research results will provide strong and sustainable technical support for Hikvision’s ongoing technology development.”The Hikvision Research Institute was established to focus on innovative product research and development and maintain the company’s status as a technology leader. Research topics include Perceptive Technology, Intelligent Analysis Technology, Big Data and Cloud Storage Technology, and Multimedia Technology. Hikvision has acquired sophisticated experience in the fields of video target detection, image segmentation, video structuring, and video retrieval. In recent years, Hikvision has participated in international competitions related to video analytics such as KITTI, MOT, ImageNet, consistently achieving remarkable performance ratings.