Huawei Cloud
The proposed architecture uses the Huawei Cloud as a platform for implementing object detection and agglomeration estimation models, as well as the Huawei Smart Cities Dashboard.
The Container Software Repository (SWR) provides a secure and reliable way of managing container images and facilitating the deployment of containerized applications. In the image below, we can see some of the containers that the team developed -- all except huwei_people_counter.
For more information on creating and uploading an image to SWR, see Software Repository for Container.
The ModelArts platform allows us easy access to data processing services, distributed training, model deployment, and much more. That way, images -- containerized applications -- uploaded to SWR can be used here.
To deploy a real-time service, we must first create an AI application. More details in Importing a Meta Model from a Container Image. Below is the first application developed by the team.
Then we must deploy the AI application created earlier in a real-time service, which can be accessed via a Rest API. More details in Deploying a Model as a Real-Time Service. Below are the applications developed by the team (except huawei-mczs-*), in particular hsc_dedicated_gpu_service.
If use is required but no GPU is available in the public resource pool, a dedicated pool can be allocated. For more details, check Resource Pools.
Object Storage Service (OBS) provides a stable, secure, efficient, and easy-to-use cloud storage service, allowing you to store virtually any volume of unstructured data in any format.
This service allowed us to store a config.json about the application deployed as a real-time service. More details about such a file in Specifications for Compiling the Model Configuration File.
However, it is no longer possible to use the file directly from OBS. More details in Object Storage Service. When creating the AI application, we must enter the settings directly in the appropriate API field.
ModelArts does not support browser requests. To solve this obstacle, the team decided to use the Huawei Cloud container engine. The Cloud Container Engine (CCE) is a scalable, high-performance container service that quickly builds clusters based on cloud servers. We can then deploy the containerized applications that we develop easily. More details at What Is Cloud Container Engine?. Below we can see the applications developed by the team -- backend and frontend (dashboard).
If you need to use GPU, you need to add the appropriate add-on: gpu-beta. Add-ons allow you to manage complementary cluster functions.
The functionality of deployed applications depends on a secure and isolated server, such as Huawei Cloud's Elastic Cloud Server (ECS). More details in What Is ECS?
