Service Flow

Figure 1 Service flow

Figure 1 shows the software service flow. For details, see Table 1.

**Table 1** Service flow description
Process	Description	Remarks
Data source	Indicates the video or image source.	The source can be RTSP streams from an IP camera (IPC) or offline videos or pictures in a disk.
Data obtaining	Implements data obtaining.	You can pull streams from the open-source FFmpeg function library or using custom implementation code. HostCPU implements data obtaining.
Data pre-processing	Implements image pre-processing functions such as decoding, scaling, and color gamut conversion.	If software decoding is used, the open-source OpenCV function library is called. The data pre-processing software runs on the host CPU or Ctrl CPU. If hardware decoding is used, the API for DVPP decoding and AIPP CSC is called to implement pre-processing. Then the corresponding hardware modules are started on the device side.
Inference	Implements the model inference function.	The function runs on the device-side AI Core or AI CPU.
Post-processing	Implements post-processing of model inference results.	The function runs on the device-side Ctrl CPU or host CPU.
Output	Displays the implementation result.	-