Data Compression Standards Committee
The IEEE Data Compression Standards Committee (DCSC) is chartered by the IEEE Computer Society. It is responsible for managing the developing of standards within the technical area of data compression and its associated applications, including without limitation data compression algorithms, data compression metrics, transmission methods, and cybersecurity issues related to data compression. It now has 7 Standard working groups.
Active Standards and Current Projects
IEEE 1857™-2023
Title: Standard for Advanced Audio and Video Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2023-6-5
Need of the project: This project is a revision for the existing IEEE 1857 and 1857.a standards. These standards are widely used in many applications. Based on the feedback from the applications, we need to improve the accuracy and robustness of the standard specification.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Video and audio service providers, including broadcasting operators, Internet video service providers
- Audio and visual content providers
IEEE 1857.2™-2023
Title: Standard for Advanced Audio Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2023-6-5
Need of the project: There are some alternative specifications with similar purpose but they do not satisfy the need for balance between efficiency and complexity required for providing high quality aural and visual service in limited band width settings. The committee views standardization as essential for lowering the cost of solutions intended for low-band width consumer devices.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Aural and visual content providers
- Video and audio service providers, including broadcasting operators, Internet video service providers
IEEE 1857.3™-2023
Title: Standard for System of Advanced Audio and Video Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2023-11-8
Need of the project: There are some alternative specifications with similar purpose but they do not satisfy the need for balance between efficiency and complexity required for providing high quality aural and visual service in limited band width settings. The committee views standardization as essential for lowering the cost of solutions intended for low-band width consumer devices.
Stakeholders: Audio and video product (hardware or software) manufacturers and vendors, audio and video content providers and audio and video service providers, including broadcasting operators, and Internet video service providers.
IEEE 1857.4™-2018
Title: Standard for 2nd Generation IEEE 1857 Video Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2018-10-23
Need of the project: There are some alternative specifications with similar purpose on video data compression by some organizations and companies, but they stay on the position of television broadcasting and telecommunication, their results are optimized for the demand of broadcasting and communication industry, but may not satisfy the demand of other industries such as internet and surveillance system. IEEE SA 1857 working group focuses on codec for internet and surveillance system, also covering other related industries. The working group views standardization as essential for improving the coding efficiency for high volume video data applications and low-bandwidth consumer devices.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Video and audio service providers, including broadcasting operators, Internet video service providers
- Aural and visual content providers
IEEE 1857.5™-2015
Title: Standard for Advanced Mobile Speech and Audio (published)
Chair: Siwei Ma
Approved by Standards Board: 2015-12-5
Need of the project: A highly efficient way is provided to transmit and store audio and speech data under constraints that include limited complexity and bandwidth. There are some alternative specifications with similar purpose but they do not satisfy the need for balance between low bit rate and mobile communications for providing high-quality aural services in limited bandwidth settings.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Aural and visual content providers
- Video and audio service providers, including broadcasting operators, Internet video service providers
IEEE 1857.6™-2018
Title: Standard for Digital Media Content Description (published)
Chair: Siwei Ma
Approved by Standards Board: 2018-12-5
Need of the project: There are some alternative specifications with similar purpose but they are created mainly for the broadcasting industry and do not have the scalability of the content descriptions to satisfy the need for large scale multimedia data searching and mobile computing.
Stakeholders:Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Video and audio service providers, including broadcasting operators, Internet video service providers
- Aural and visual content providers
IEEE 1857.7™-2018
Title: Standard for Adaptive Streaming (published)
Chair: Siwei Ma
Approved by Standards Board: 2018-12-5
Need of the project: HTTP streaming has become a popular approach for delivering of multimedia content over Internet. A standard for HTTP streaming of multimedia content would allow a standard-based client to stream content from any standard-based server, thereby enabling interoperability required by the market and the industry of Internet Streaming.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Media content creators
- Media content service providers, including broadcasting operators, Internet video service providers
- Advertisement service providers
IEEE 1857.8™-2020
Title: Standard for 2nd Generation Audio Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2020-6-4
Need of the project: There are some alternative specifications with similar purpose but they do not satisfy the need for balance between efficiency and complexity required for providing high quality aural and visual service in limited band width settings. The committee views standardization as essential for lowering the cost of solutions intended for low-band width consumer devices.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Aural and visual content providers
- Audio and video service providers, including broadcasting operators, Internet audio or video service providers
IEEE 1857.9™-2021
Title: Standard for Immersive Visual Content Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2021-11-9
Need of the project: Today’s TV and Internet video provide a same view of content to all users. But a new movement just appears from internet and new media industry, they want to provide an immersive content display to end user, such as supporting zoom in and out, change view position and angle, and so on, controlled by end user. The immersive visual content includes but not limited to panorama video, free view video, light field, 3D model and other kinds of synthetic visual contents. It is no doubt that the data quantity of immersive visual content is much higher than that of traditional 2D/3D video, so it might take more than 10 times bandwidth cost for supporting this kind of applications. However, and there is no specification exist for the this special purpose to compress the high volume immersive visual content. The committee views standardization as essential for providing high efficiency coding and unified format for immersive visual content based applications.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Video and audio service providers, including broadcasting operators, Internet video service providers
- Aural and visual content providers
IEEE 1857.10™-2021
Title: Standard for Third Generation Video Coding (published)
Chair: Siwei Ma
Approved by Standards Board: 2021-11-9
Need of the project: There are some alternative specifications with similar purpose but they are not able to satisfy the demand of higher coding efficiency for video data, especially for the surging ultra-high definition videos (UHD), i.e. 4K and 8K, and VR (Virtual Reality) video. The committee views standardization as essential for improving the coding efficiency for high volume video data applications and low-bandwidth consumer devices.
Stakeholders: Stakeholders being benefited include but are not limited to:
- Audio and video products (hardware or software) manufacturers or vendors
- Video and audio service providers, including broadcasting operators, Internet video service providers
- Aural and visual content providers
P1857.11
Title: Standard for Neural Network-Based Image Coding (on going)
Chair: Siwei Ma
PAR Approved by Standards Board: 2021-9-23
Need of the project: There are some alternative specifications with similar purpose but they do not satisfy the need for balance between efficiency and complexity required for providing high quality aural and visual service in limited band width settings. The committee views standardization as essential for lowering the cost of solutions intended for low-band width consumer devices.
Stakeholders: Stakeholders include:
- Imaging products (hardware or software) manufacturers or vendors
- Image-enabled service providers, especially Internet image-enabled service providers
- Neural network-oriented computing products (hardware or software) manufacturers or vendors
- Visual content providers
IEEE 1857.12™-2022
Title: IEEE Standard for Smart Media Transport (published)
Chair: Siwei Ma
Approved by Standards Board: 2022-11-08
Need of the project: Various contents are required to be organized flexibly. Different networks, e.g., broadcasting network, broadband network, and mobile network, can be used for media transport. The presentation terminals are various. A common approach for media encapsulation, delivery and presentation is required. This standard specifies technologies for the common encapsulation, adaptive transmission, and multi-screen presentation.
Stakeholders: Stakeholders include but are not limited to
- Media content creators
- Media content service providers, including broadcasting operators, Internet video service providers
- Advertisement service providers
IEEE Std 2941™-2021
Title: IEEE Standard for Artificial Intelligence (AI) Model Representation, Compression, Distribution, and Management (published)
Chair: Yonghong Tian
Approved by Standards Board: 2021-12-08
Need of the project: The standard satisfies the demand of the unified representation of heterogeneous computing platforms and frameworks, higher efficiency of inference and compression, and secure and credible distribution and management of models, especially on resource-constrained devices and large-scale model distribution learning. Standardization is essential for improving the efficiency of large-scale AI model distribution management and application and resource-constrained devices.
Stakeholders: Industry and commerce: AI products (hardware or software) manufacturers or vendors, AI service providers, AI model providers; Government; Consumers; Academic and research bodies; Standards application businesses
IEEE 2941.1™-2022
Title: IEEE Standard for Operator Interfaces of Artificial Intelligence (published)
Chair: Yonghong Tian
Approved by Standards Board: 2022-11-08
Need of the project: The great abundance of artificial intelligence (AI) software and hardware, though providing many conveniences for practical usage, bring increasingly more challenges of diversification, complexity and fragmentation. The complicated mapping between AI hardware and software gradually become a major obstacle to the efficient deployment of AI-related applications. By defining a series of standardized operator interfaces, this standard can not only help AI software practitioners quickly adapt their software with various AI processors, but also assist AI hardware developers to efficiently integrate their hardware into the existing AI software ecology.
Stakeholders: AI hardware vendors, AI software developers, AI service providers, government regulators, consumers, academics and researchers, standards application businesses.
IEEE 2941.2™-2023
Title: IEEE Standard for Application Programming Interfaces (APIs) for Deep Learning (DL) Inference Engines (published)
Chair: Yonghong Tian
Approved by Standards Board: 2023-09-21
Need of the project: With the great abundance of deep learning (DL) applications, a variety of inference engines emerged over the past few years. Each DL inference engine defines its own APIs, which may lead to industrial fragmentation and introduce complexity for developers. In order to solve the above issues and further promote the application of DL technology, a standard for application programming interface (API) of deep learning inference engines is needed.
Stakeholders: Deep learning inference engine manufacturers or vendors, deep learning application and service providers, academic and research bodies, standards application businesses.
P2941.3
Title: Standard for Representation and Application Programming Interface (API) of Large-scale Pre-trained Artificial Intelligence (AI) Models (on going)
Chair: Yonghong Tian
PAR Approved by Standards Board: 2023-06-05
Need of the project: Large pre-trained models are mainly based on the Transformer structure, which needs a larger number of parameters and higher computational consumption. The standard specifies a complete process of training, inference, transfer, mixture, instruct-tuning, representation, compression, distribution, services, etc. This project provides a pipeline for developing and applying large-scale pre-trained AI models based on this standard.
Stakeholders:
- Industry and commerce: AI products (hardware or software) manufacturers or vendors, AI service providers, large-scale pre-trained model providers.
- Government
- Consumers
- Academic and research bodies
- Standards application businesses
P3122
Title: Standard for Data Processing and Compression Framework for Internet of Things (on going)
Chair: Jinghui Lu
PAR Approved by Standards Board: 2021-11-09
Need of the project: The standard responds to the demand of the information exchanges and application data integrations for IOT products. There are many kinds of information transmitted from IoT front end devices, such as, camera, smart bracelet, and sensor, to IoT center node where the IoT data are collected, stored, and analyzed. The IoT data includes video, audio, digital label or characteristic data, sensor data, etc. The standard provides the framework and tools to improve the efficiency of transmission and storage, protect the security and integrity of data.
Stakeholders: Industry and commerce: IoT products (hardware or software) manufacturers or vendors; Consumers
IEEE Std 3161™-2022
Title:IEEE Standard for Digital Retina Systems (published)
Chair: Yaowei Wang
Approved by Standards Board: 2022-12-03
Need of the project: Ubiquitous camera networks in current smart cities create a massive amount of images and videos at a range of spatial-temporal scales. However, the capabilities of sensing systems often lag behind the fast growth of video applications. Therefore, digital retina systems, which provide a novel visual computing framework, are designed to align high-efficiency sensing models with video coding, feature coding, model coding, as well as their joint optimization. In particular, the compressed video stream targets for human vision, the compact feature stream for machine vision, and the model stream incrementally update deep learning models to improve the performance of human/machine vision tasks. Most of the existing technology standards define a one-camera-one-stream framework, whereas, digital retina systems employ a one-camera-three-streams framework. Consequently, digital retina systems allow comprehensive, intelligent, and efficient interactions between retina-like cameras and edge servers over the cloud through the three streams. Digital retina systems can also achieve a higher compression ratio for visual data while maintaining competitive performance with uncompressed signals for various visual analysis tasks. They can support precise localization and tracking of objects across multiple cameras in different cities, playing a fundamental role in visual big data analysis and retrieval in smart cities. Currently, there is no existing specification for a visual computing architecture based upon the three streams.
Stakeholders: Industry and commerce: AI chip manufacturer, equipment manufacturers, AI algorithm or service providers, edge computing vendors, cloud computing vendors, and multimedia related vendors.
P3161.2
Title: Standard for End Subsystem of Digital Retina Systems (on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-09-21
Need of the project: With the development of Artificial Intelligence (AI) technology, the number of visual end-devices with AI capabilities is increasing in many fields, such as intelligent transportation, intelligent manufacturing and smart city. However, the traditional visual end-device is designed with the one-camera-one-stream framework, and only produces a compressed video stream that cannot be directly used for data analysis and retrieval. The end subsystem of DRS produces both the feature stream for machine vision and the compressed video stream for human vision. Moreover, end subsystem functions can be defined to adapt to diverse application scenarios. There are no available standards to satisfy the above requirements.
Stakeholders: the system architects, developers and users of end devices.
P3161.3
Title: Standard for Edge Subsystem of Digital Retina Systems (on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-09-21
Need of the project: With the widespread distribution of visual end-devices, edge devices are deployed to satisfy the real-time processing demands. Traditional edge systems are designed based on the one-camera-one-stream framework, which primarily receive and process the video stream. Instead, the edge subsystem of DRS addresses the economic processing of video stream, feature stream and model stream. It provides functionalities of multi-channel data aggregation and forwarding, cooperative resource scheduling, and data computing. It receives the model stream for inference and model training in edge side. Moreover, edge subsystem functions can be re-defined to adapt to diverse application scenarios. There are no available standards to satisfy the above requirements.
Stakeholders: The system architects, manufacturers of edge devices, developers and users of edge services
P3161.4
Title: Standard for Cloud Subsystem of Digital Retina Systems (on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-09-21
Need of the project: Unlike existing cloud systems, the cloud subsystem of DRS shall simultaneously process video, feature, and model streams due to the characteristics of one-camera-three-streams framework of DRS. The cloud subsystem of DRS can both manage and process video, feature, and model streams, collaboratively. Moreover, it trains, disseminates, and updates large-scale models. It provides functionalities of collaborative interaction, data aggregation, storage, mining, and decision making at the global level. Therefore, standardization of cloud subsystem is essential to support global efficient coordination of DRS.
Stakeholders: Industry and commerce: equipment manufacturers, visual algorithm or service providers, cloud computing providers, manufacturers of cloud servers, users of cloud service, and multimedia-related vendors.
P3161.5
Title: Standard for Algorithm and Model Repository of Digital Retina Systems (on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-11-08
Need of the project: In traditional video-related applications, data collected from end devices is transmitted to cloud servers for analysis and processing. However, DRS has the characteristics of software-definability and model updating mechanism. Algorithms and models can be deployed and updated on end, edge, and cloud devices with heterogeneity to generate video, feature, and model streams. Algorithms also need to collaborate to accomplish data analysis tasks. Therefore, it is necessary to provide a unified description and management of algorithms with diverse functionalities, in order to support the operation and collaboration of algorithms within heterogeneous end devices, edge devices, and cloud servers in DRS. Currently, there are no corresponding standards that can fulfill this requirement.
Stakeholders: Industry and commerce: AI chip manufacturers, equipment manufacturers, AI algorithm or service providers, edge computing vendors, cloud service vendors, and multimedia-related vendors.
P3161.6
Title: Standard for Storage System of Digital Retina Systems(on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-11-08
Need of the project: DRS can be used for the large-scale applications in fields such as smart transportation, public safety, and smart cities. Thus, massive amounts of heterogeneous data including video, feature, analysis result carrying unified temporal-spatial identification, algorithms and models, should be stored properly in DRS. Furthermore, DRS is designed to realize the collaborative interaction among storage services by video fetching on demand and storage resource allocation. Currently, there is no available standard to meet the requirements of storage system for DRS. Therefore, the standardization of storage system will effectively enhance the capabilities of DRS.
Stakeholders: Industry and commerce: equipment manufacturers, storage devices vendors, system integrators, edge computing vendors, cloud service providers.
P3161.7
Title: Standard for End-edge-cloud Collaboration of Digital Retina Systems (on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-11-08
Need of the project: Visual tasks are characterized by various types and a large amount of computation. The objective of end-edge-cloud collaboration is to improve the utilization of system resources and efficiency of visual data processing. Within the DRS framework, multiple data streams including video, feature and model are leveraged for collaboratively processing and understanding the visual information across the end-edge-cloud platform, to enable real-time gathering of valuable information and collaborative data analysis. Currently, there is no corresponding standard which can meet the needs of these characteristics. Therefore, standardization of end-edge-cloud collaboration of DRS is essential to establish corresponding specifications which tends to make better use of the system and further reduce the computing cost.
Stakeholders: Industry and commerce: AI chip manufacturers, equipment manufacturers, AI algorithm or service providers, edge computing vendors, cloud computing vendors, and multimedia-related vendors.
P3161.8
Title: Standard for Security and Privacy Protection of Digital Retina Systems (on going)
Chair: Yaowei Wang
PAR Approved by Standards Board: 2023-09-21
Need of the project: Although standard development organizations, such as ISO and IEC, have developed security standards for end and cloud systems, they are not entirely applicable to DRS due to its collaborative security strategy with the end, edge, and cloud subsystems, as well as its distinct data processing needs for video, feature, and result streams containing privacy-relevant data. It is necessary to build a security and privacy protection strategy applicable to DRS as a whole to facilitate the product design of all stakeholders. A standard is necessary to set the security and privacy protection requirements of DRS in order to help ensure the systems’ availability, confidentiality, and integrity as well as to satisfy privacy protection requirements.
Stakeholders: Industry and commerce: equipment manufacturers, AI algorithm or data service providers, edge computing vendors, cloud computing vendors, multimedia-related vendors, designers, developers, system integrators, testers.
IEEE 3161.9™-2023
Title: IEEE Standard for Protocols and Interfaces of Digital Retina Systems (published)
Chair: Yaowei Wang
Approved by Standards Board: 2023-09-21
Need of the project: Unlike existing visual systems with the one-camera-one-stream framework, DRS has a one-camera-three-streams framework, in which multiple data streams such as video, feature, and model streams are generated and transmitted in parallel. Currently, there are no corresponding standards that are able to satisfy the transmission requirements within this framework. Therefore, standardization of Protocols and Interfaces of DRS is essential to realize effective interconnection and interaction among multi-source heterogeneous end devices, edge servers and cloud center in the system.
Stakeholders: Industry and commerce: AI chip manufacturers, equipment manufacturers, AI algorithm or service providers, edge computing vendors, cloud computing vendors, and multimedia related vendors.
P3184
Title: Standard for Data Framework for Autonomous Driving (on going)
Chair: Yanyong Zhang
PAR Approved by Standards Board: 2022-06-16
Need of the project: A standard is needed to define a data-related architecture, to enable the rapid development of highly efficient autonomous driving systems and support growth of the autonomous driving industry. Autonomous driving systems include multiple subsystems such as sensors, actuators, machine learning models, heterogeneous computing platforms, and vehicle-road-cloud collaborative control components. These subsystems impact the level of effective data interaction and data processing that a system can achieve. Today, the data frameworks (data format, data processing flow, data interaction protocol) of all these subsystems is designed individually, which severely limits the efficiency of information exchange among the various subsystems. The lack of common interfaces hinders the fast development of autonomous-driving-related perception, decision-making, and planning algorithms.
Stakeholders: Autonomous vehicle manufacturers, AI chip manufacturers, sensor and communication product manufacturers, AI algorithm developers, service providers, edge computing vendors, cloud computing vendors, and multimedia vendors.
P3184.1
Title: Standard for Joint Calibration Framework of Multiple Roadside Sensors (on going)
Chair: Yanyong Zhang
PAR Approved by Standards Board: 2023-06-05
Need of the project: The joint calibration of multiple roadside sensors is mainly required by traffic management departments, intelligent transportation system builders, road facility management departments and other institutions and organizations that need real-time and accurate traffic monitoring. These departments and organizations need to obtain accurate and consistent calibration results to support real-time traffic monitoring, road planning, traffic safety and other applications. At the same time, joint calibration is also adopted by traffic data analysts and researchers to analyze the defects of calibration results, explore some methods to obtain more accurate and reliable traffic data, and support higher quality and more precise traffic analysis and decision-making.
Stakeholders: Vehicle manufacturers, AI chip manufacturers, equipment (sensors, communication products etc.) manufacturers, AI algorithm or service providers, edge computing vendors, cloud computing vendors, and multimedia related vendors.
P3184.2
Title: Standard for Property Graph Description of Road Networks (on going)
Chair: Yanyong Zhang
PAR Approved by Standards Board: 2023-06-05
Need of the project: With the increasing popularity of intelligent transportation systems and other advanced applications, there is a growing demand for high-quality road network data. The road network describes the important information of the spatial structure and topological relationship of roads and streets. The graph format provides the high efficiency and rich query capabilities in data storage and retrieval, which can provide fast and scalable storage and access for applications such as autonomous driving simulation, scene database construction and use. It can also be needed by urban planners and engineers to design and optimize transportation systems. Additionally, emergency responders can use it to optimize response routes, and traffic data flow analysts can use it to make traffic predictions.
Stakeholders: Vehicle manufacturers, AI chip manufacturers, equipment (sensors, communication products etc.) manufacturers, AI algorithm or service providers, edge computing vendors, cloud computing vendors, and multimedia related vendors.
P3366.1
Title: Standard for Geometry Point Cloud Compression (on going)
Chair:Shan Liu
PAR Approved by Standards Board: 2023-06-05
Need of the project: With the advent of technology, the demand for more immersive experience beyond the conventional 2-D video viewing continues to rise. Point cloud is a data format which can be used to form 3-D models and provide a six-degree-of-freedom viewing experience. A point cloud usually consists of huge amount of data, thus efficient compression is critical for point cloud to be used in many applications. Standardization of point cloud compression is essential for transmission of high volume point cloud data over bandwidth constrained networks and processing such data on storage constrained consumer devices.
Stakeholders:
- Manufacturers and vendors (hardware or software) with products using point clouds
- Service providers, including broadcasting operators, Internet service providers distributing point clouds or models developed from point cloud data
- Point cloud or point cloud generated content providers
P3366.2
Title: Standard for Polygonal Mesh Compression (on going)
Chair:Shan Liu
PAR Approved by Standards Board: 2023-06-05
Need of the project: With the advent of technology, the demand for more immersive experience beyond the conventional 2-D video viewing continues to rise. Mesh is a data format which can be used to form 3-D models and provide six-degree-of-freedom viewing experience. A mesh model usually consists of huge amount of data, thus efficient compression is critical for mesh models to be used in many applications. Standardization of mesh compression is essential for transmission of high volume mesh data over bandwidth constrained networks and processing such data on storage constrained consumer devices.
Stakeholders: – Manufacturers and vendors (hardware or software) with products using mesh data
- Service providers, including broadcasting operators, Internet service providers distributing mesh models or visual effect developed from mesh models
- Mesh or mesh generated content providers
P3404
Title: Standard for Requirements and Framework for Sharing Data and Models for Artificial Intelligence across Multiple Computing Centers (on going)
Chair: Yue Yu
PAR Approved by Standards Board: 2023-09-21
Need of the project: With the emergence of technologies such as large-scale distributed machine learning, Artificial Intelligence (AI) for science, and joint-cloud computing, the need of sharing the raw data, intermediate computing results, and model parameters across geographically distributed computing resources also increases rapidly. However, the data and model to be shared are usually heterogeneous, with large sizes, and have different privacy requirements, which hinders the feasibility and efficiency of data and model sharing across multiple computing centers. Thus, it is essential to identify the new requirements and standardize the framework of data and model sharing procedures across multiple computing centers such that the interconnection and collaboration of computing centers can be exploited in real-world applications.
Stakeholders: Industry and commerce: AI products (hardware or software) manufacturers or vendors, AI service providers, AI model providers, equipment manufacturers, system developers, computing center owners.
P3483
Title: Standard for Deep Intermediate Feature Coding (on going)
Chair: Lingyu Duan
PAR Approved by Standards Board: 2024-05-22
Need of the project: With the edge and cloud computing technologies becoming popular, in pursuit of analytical efficiency, the proportion of features in all transmission media is gradually increasing. Deep intermediate feature compression serves as the foundation for efficient feature computation in the context of front-cloud collaboration. Currently, the rapid development of large foundation models and collaborative infrastructure poses significant challenges to existing standardized efforts and efficient perception paradigms. In this context, deep intermediate feature compression, as an important tool that significantly improves the efficiency of front-cloud collaboration, holds the potential to adapt to a wide range of network architectures. Its standardization will facilitate the rapid development of intelligent applications in front-cloud architectures. There is no alternative specification with a similar function.
Stakeholders: Stakeholders include but not limited to:
- Intelligent analytics products (hardware or software) manufacturers or vendors
- Surveillance content providers in smart cities
- Intelligent analytics service providers, including but not limited to intelligent transportation, smart security, autonomous driving
- Software and hardware architect in the field of intelligent analytics
