Page 223 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

(continued)

Domain Industry, Innovation, and Infrastructure
Model Training and fine-tuning • AIGC technology is proposed for the production of
training data of small sample to improve AI recognition 52-China Mobile
ability by using large models to solve problems such as
insufficient sample size and data skew in the traditional
visual AI training process.
• Based on the open-source target detection model,
secondary training is carried out to label small targets
in images, and the location and category of targets
are usually marked with bounding box. The annotation
results are saved in VOC or COCO formats.

Case Studies Computer network fusion video brain

Testbeds or pilot deployments The pilot is deployed on China Mobile's internal network.

52�2� Use case description

52�2�1 Description

Introduction: Guangdong company innovates to create computer and network integration
video intelligence brain, and carries out "artificial intelligence + video" action to promote
industrial upgrading and improve the quality of life. The platform sinks the video decoding
frame extraction and AI inference service computing power to the city node, realizing the
optimal and intelligent scheduling of video analysis computing resources at the provincial
side, the city edge side, and the user side, effectively saving 60% of bandwidth resources and
increasing the delay by 30%. Moreover, by deploying large models in the cloud, secondary
verification of recognition results is carried out to improve the accuracy of video intelligent
recognition. At the same time, in terms of data, the introduction of AIGC technology drives
the production of training data of small sample AI recognition ability by using large models
to solve problems such as insufficient sample size and data skew in the traditional visual AI
training process.

The project has landed in urban management and public safety and other fields, building smart
transportation, smart city and other business scenarios, to provide more comfortable living
conditions for urban residents; As well as industrial manufacturing and food production and
other fields, optimize the production process, efficiently supervise the production environment
and production quality and other factors, accelerate the industrialization process and improve
food safety. Subsequently, it can be extended to all walks of life to provide new quality
productivity for the development of the industry.

UN Goals:
• Goal 3: Good health and well-being
• Goal 9: Industry, Innovation, and Infrastructure
• Goal 11: Sustainable cities and communities

Justify UN Goals selection: This project has created a computing network integration of video
intelligence, a one-stop video AI product enabling system as the design concept, through
"one cloud, two libraries, three centers" as the infrastructure, with "platform and equipment",

207