Page 290 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Figure 1: AI Intelligent Middle Platform “1+1+3+N” Architecture Diagram

• 1: An integrated platform for annotation, training, and inference
The integrated platform for annotation, training, and inference, as a one-stop platform
for annotation, training, and reasoning, can achieve five unifications of computing power,
data, development, models, and capability usage. It supports the development of
algorithms through methods such as managing different computing power, (eg., NVIDIA
and Huawei) and visual drag-and-drop interfaces, covering the entire lifecycle workflow
of AI model production, thereby lowering the R&D threshold [4].
• 2: MaaS model services covering images, text, and videos
Image Services: There are 16 OCR services, including ID card recognition, social security
card image recognition, driver's license recognition, vehicle registration certificate
(driving license) recognition, train ticket recognition, etc.
Video Services [3]: There are 47 services, including vehicle detection and attribute
recognition, non-motor vehicle detection, vehicle traffic flow analysis, person detection
and attribute recognition, face detection and attribute recognition, pedestrian traffic flow
analysis, crowd gathering detection, head and shoulder hard hat detection for personnel,
etc.
Text Services: There are 17 services, including text dialogue emotion recognition,
sensitive word detection in text, short text similarity, entity extraction, and keyword
extraction, among others.
• 3: AI tool platforms
Large Model and Agent Aggregation Platform serves as a connection layer between
general-purpose large models and industry applications, aggregating various open-
source large models, managing industry data knowledge bases, plugin tools, workflows,
and other components, supporting low-code, visual rapid construction of large model

254