The nine major projects jointly build an open source and open ecosystem.

At the 2022 World Artificial Intelligence Congress (WAIC) Science Studies held today, the Shanghai Artificial Intelligence Laboratory released the "OpenXLAB Puyuan" artificial intelligence open source open system. The system includes nine open source projects, covering from perception to decision -making, from plane to three -dimensional, from data to calculation, from technology to human education.

Lin Dahua, a professor at the Shanghai Artificial Intelligence Laboratory, said: "X represents unknown, open and growth, and 'Puyuan' means that the Shanghai Artificial Intelligence Laboratory is based on the banks of the Huangpu River. Artificial intelligence open source technology, together with partners to build an artificial intelligence open source open ecology. "It will promote the comprehensive empowerment of artificial intelligence in various fields of industry, university, and research, and to build an artificial intelligence open ecology, promote artificial intelligence research and technological breakthroughs, cross -innovation and industrial industries The landing provides a full range of algorithms and platform support.

New upgrade, bringing new architecture, new algorithms, new ecology

The newly upgraded "OpenXLAB Puyuan" artificial intelligence open source open system includes: OpenMMLAB 2.0 Pu visual visual algorithm open source system, OpenGVLAB school universal visual open system, OpenDilab 1.0 Pura decision intelligent open source platform, and the first appearance of OpenXDLAB Puppet high -quality figures Content platform, OpenXRLAB Po Po expanded reality open source platform, OpenDatalab Puku Puci artificial intelligence open data platform, and OpenComputelab Puku Puku artificial intelligence open computing system.

As the most influential visual algorithm open source system in the era of deep learning, OpenMMLAB 1.0 Since its release in July last year, various influence indicators have increased significantly, and have been used by millions of developers in more than 100 countries and regions around the world.

The newly upgraded OpenMMLAB Paradist 2.0 visual algorithm open source system introduces a new training architecture Mmentine, which realizes the unity of interfaces of different code libraries, making the expansion of the new algorithm more flexible and efficient. At the same time, we build 6 new algorithm libraries around the front -edge direction of self -supervision and learning to achieve seamless connection from model training, deployment to reasoning, and open up the last mile of AI landing, and widely adapt to training and reasoning chips. Common ecological development.

From perception to decision -making, continue to move towards GM artificial intelligence

OpenGVLAB scholar universal visual open source open platform has released high -efficiency pre -training models for academic and industry circles to achieve classification, target detection, semantic segmentation, and depth estimation of four major tasks. Openai's CLIP model. At present, the OpenGVLAB scholar 2.0 is creating a universal visual technology system covering data, models, evaluations, standards and applications, etc. to help developers quickly reduce the threshold for development of general visual models.

Decision intelligence is the key area of ​​artificial intelligence development. OpenDILAB Poor decision -making intelligent open source platform, as the world's first decision -making AI platform covering the most complete academic algorithm and industrial scale, has opened up the closed loop of decision -making AI research and industrial needs from the bottom up.

From the plane to the three -dimensional, build a bridge connecting the real world and the virtual world

Under the support of technologies such as 3D vision and virtual reality, the Yuan universe has become a new industrial form, and data content is the core element of the development of the Yuan universe. OpenXDLAB Pu Popia high -quality digital content platform provides a large amount of high -quality dynamic three -dimensional data, covering multi -perspective videos and three -dimensional models such as face, human and objects and other categories. The platform also provides a powerful content generating engine, which can combine the scene and characters to renders a rich and realistic dynamic three -dimensional scene, which greatly improves productivity.

In addition to the rich digital content, the laboratory also jointly created OpenXRLAB Puxi expansion reality open source platform with Zhejiang University and Shangtang Technology. The perception ability of the real world has been increased from a two -dimensional plane to a three -dimensional space; the XR multi -mode man -machine interaction platform provides human -machine interaction capabilities in 3D space; XR rendering generation platform provides the final rendering capabilities and builds out. Bridge connecting the real and virtual world.

From data to calculation, consolidate artificial intelligence innovation cornerstone

The rapid progress of artificial intelligence technology is inseparable from the two important cornerstones of data and computing power. The new OpenDatalab Pu Po Po Po number artificial intelligence open data platform brings together a large amount of data resources, including more than 3,000 data sets covering more than 800 tasks, and providing convenient search and fast download services. OpenDatalab is also closely connected with the open source algorithm system of OpenXLAB. Through a simple command line tool, one -click deployment and out of the box can be used.

With the rapid evolution of algorithms, various architecture chips have emerged endlessly, and the adaptation of algorithms and chips has become one of the most important challenges in the field of artificial intelligence computing today. OpenComputelab Puku's artificial intelligence open computing system aims to make AI computing more open, more efficient and simpler. It took the lead in launching the "Occupible Map Labram", from the perspective of the application of algorithms, systematically sorted out hundreds of more than hundreds of operators related to deep learning, formed a classification and grading system, and built an open evaluation framework, which can calculate performance performance. Carry out a comprehensive and in -depth analysis to provide valuable reference guidelines for the collaborative evolution of upstream and downstream. From science and technology to human education, realize AI sustainable development

Artificial intelligence is closely related to the ethics and governance of human society. The laboratory coordinated Tsinghua University and Fudan University to jointly create an open innovation platform for OpenGlab dandelion artificial intelligence governance. With the meaning of hope, warmth, and friendship, the platform aims to build a system, practical artificial intelligence governance infrastructure, explore the "rules-technology-scene-evaluation" integrated governance innovation system, promote multidisciplinary cross-interdisciplinary, diverse subjects Share and co -construction to promote artificial intelligence credible and sustainable development.

About Shanghai Artificial Intelligence Laboratory

Shanghai Artificial Intelligence Laboratory is a new scientific research institution in the field of artificial intelligence in my country. It carried out strategic, original, and forward -looking scientific research and technical research, breaking the important basic theory and key core technologies of artificial intelligence, and creating a "breakthrough, leading type, leading type , Platform -type "integrated large -scale comprehensive research base, supporting my country's artificial intelligence industry to achieve leapfrog development, targeting the internationally first -class artificial intelligence laboratory, and becoming a world -renowned source of artificial intelligence original theory and technology.

Author: Shen Yisha

Editor in charge: Ren Quan

- END -