Introduction

Kneron PLUS stands for Platform Library Unified Software which is a framework comprising new software(SW) and firmware(FW) design for KL520 (and alpha for KL720).

In order to run the inference of models on Kneron AI dongles, there are three parts of AI application development are required:

Below diagram depicts three parts of development in a big picture.

However, this document only focuses on the software development and the firmware development. For the model development, please refer to the Toolchain Docker part.

In comparison with the previous SW/FW framework, this aims to simplify the design flow for AI applications development.

Below gives some definitions regarding the Kneron PLUS:

The features which PLUS Supported are listed below:

Index Category Supported Item KL520 KL720
1 Image Format RGBA8888 , RAW8 , YUYV422 , RGB565
2 System Firmware In Flash
3 Model In Flash
4 Runtime Upload Firmware X
5 Runtime Upload Model
6 Software Reset
7 Scan Devices
8 Device Log via USB
9 Device Connection All Dongles, Specified Dongle(s)
10 Inference Sync / Async Mode Inference
11 Multiple Dongle Auto Dispatch
12 Enable / Disable Pre-process on Device
13 Enable / Disable Post-process on Device
14 Output Floating Point Result
15 System / Model Info Get Firmware Version
16 Get KN Number
17 Get Model CRC
18 Get Model Info
19 Application API App Yolo Inference
20 Generic Inference
21 Customized Inference (C code only)
22 Inference Examples App Yolo Inference (Single Thread)
23 App Yolo Inference (Multiple Threads)
24 App Yolo Inference (Model in Flash)
25 App Yolo Inference (Web Cam)
26 App Yolo Inference (Web Cam with Drop Frame)
27 Generic Inference (Raw Output)
28 Generic Inference (with Crop)
29 Generic Inference (with Post Process on Host Side)
30 Customized Inference with Single Model (C code only)
31 Customized Inference with Multiple Models (C code only) X
31 Model Zoo Examples Simple examples for pre-trained models

The following components are contained in Kneron PLUS: