In order to run customized models on Kneron AI dongle, there are four stages are involved:
This will not be introduced in this document.
The software interface for sending requests to firmware and receiving results from firmware.
SCPU Firmware Development
The entry of the firmware. Which models should be run on NCPU Firmware and in what sequence should these models run are determined in SCPU Firmware.
NCPU Firmware Development
Where models actually run. Besides the model inference, the preprocess and the postprocess can be chosen to run on NCPU Firmware.
The diagram below demostrates the inference flow for every models runnig on Kneron AI dongle, and how the PLUS, SCPU, and NCPU interact with each other.