亚信科技

A few days ago, the 1st Global AI Innovation Contest, founded by Chinese Association for Artificial Intelligence and other units, came to an end in Hangzhou. After multiple rounds of fierce competition, Telecom AI Labs of AsiaInfo stood out from 4,256 participants in the world, ranking 17th to win the third prize of the contest.

Oriented by the international community, this contest is open to the world, focusing on AI technology and application innovation. Three contest tracks of “Medical Image Report Anomaly Detection”, “PANDA Large Scene Multi-Object Detection and Tracking” and “Semantic Matching of Dialogue Short Text” have been set up, attracting a total of more than 10,000 teams from 21 countries to participate in the contest. In the case that none of the three tracks is the main direction of Telecom AI Labs of AsiaInfo, the participating team still won the 17th place in the contest of “PANDA Large Scene Multi-Object Detection and Tracking” through competing with 4,255 teams from all over the world, finally getting the third prize.

Award certificate

This competition aims to promote the maturity and application of large-scale, long-range, and multi-objective visual analysis technology. Relying on its own product development and project experience, AsiaInfo’s participating team started from the perspectives of timeliness, accuracy, and portability, conducted multi-scale feature extraction and multiple-layer fusion of Re-ID information through parallel sliding window slicing and network structure reconstruction, and integrated model acceleration and containerization technology, indecently researching and forming a set of “end-to-end” target detection and tracking algorithm system.

PANDA data

Model structure diagram

Judging from the contest data, the cloud-native container technology and end-to-end model algorithm structure are stable, and the overall scheme is 17.2% higher than the open-source baseline performance. They have obvious advantages in terms of speed and accuracy in dynamic image multi-target detection and tracking scenes, and the scheme can be well implemented. At present, the model algorithm has been packaged for productization, and integrated in the AISWare AI² Edge of AsiaInfo.

AISWare AI² Edge is an intelligent edge-side product that integrates software and hardware with AI analysis capabilities for customers to meet the needs of security and production management for people’s livelihood. It can analyze video content in real time, automatically detect abnormal information, and proactively prevent and control risks, and make the production, life, operation and management more intelligent in the scenes including communities, parks, buildings, schools, shopping malls, hospitals, factories, construction sites, mines, and power plants. The product introduces AsiaInfo’s universal artificial intelligence capabilities, and adopts a new distributed parallel computing architecture to scale horizontally to thousands of nodes. It has powerful computing and processing capabilities and is “out-of-the-box” for simple and reliable use, which has been widely used in the scenes such as smart shopping malls, smart parks, smart construction sites, smart computer rooms, smart power plants, and smart elderly care. AISWare AI² Edge and its solutions won the “CIS Annual Innovative Product Award” and “Excellent Solution of China’s Software Industry” in 2020.

Telecom AI./ML. Labs of AsiaInfo

Established in 2019, the labs are working on R&D, innovation and industrialization in the cross-field of communications and artificial intelligence, and it continues to improve AsiaInfo’s industry influence in the field of data science and telecom artificial intelligence through in-depth exchanges and cooperation with domestic and foreign industry, academia and research institutions. The labs keep up with market demand and promote AI technology innovation and product R&D, and have accumulated leading technologies and complete solutions in the fields of 5G intelligence, AIOps, visual analysis, and OCR. The labs have applied for more than 100 international and domestic patents; formulated over ten international and domestic standards for 3GPP, GSMA, ITU, etc.; obtained a number of AI software copyrights; and achieved good results in many international AI contests. All members of the labs come from well-known universities at home and abroad, with members having master degree or above accounting for 100%. A joint laboratory was established with Institute for AI Industry Research, Tsinghua University to realize the sharing and joint development of high-end AI talents and capabilities. Through industry-university-research cooperation, an integrated operation model of AI algorithm research and development, operation, and service has been formed, which adds and integrates intelligence into AsiaInfo’s global products in an all-round way.