Development Workflow

Process Description

Figure 1 API development process

Key Steps

1. Initialize Vision SDK APIs before using them to develop applications.
2. Perform media data processing/model inference.
   1. Media data processing includes image encoding and decoding, cropping, resizing, and padding.
   2. Model inference, including object recognition and image classification, consists of the following steps.
      1. Before model inference, prepare a proper OM model. You can also convert other pre-trained models into OM models. For details, see CANN ATC Instructions.
      2. Load the prepared model to the system as a file or through memory.
      3. Optional: Perform media data processing. Before model inference, operations can be performed on the input media data, including image decoding, cropping, resizing, and padding.
      4. Run model inference to implement functions such as image classification and target recognition.
      5. Optional: Perform postprocessing on the data output after model inference. The inference result is processed based on user requirements. For example, it can be processed as a specific result that is readable. Model postprocessing supports the following ways:
         • (Recommended) Select a model postprocessing mode by yourself.
         • Call the Vision SDK APIs.
3. Perform deinitialization by calling mx_deinit().