How to Compile and Use Amlogic NN Api demo

Amlogic NN Api is a set of NPU Api officially launched by amlogic. This document will introduce how to compile and use khadas to demonstrate based on this set of Api.

API Docs

For detailed information about the API, please refer to the document docs/zh-cn/DDK_6.4.3_SDK_V1.6 API Description.pdf

Compile

Get demo source code

The source code of the aml_npu_nnsdk_app repository is open on the official gitlab of Khadas

$ mkdir workspace
$ cd workspace
$ git clone https://gitlab.com/khadas/tengine_khadas_app.git

Source code structure introduction

There are currently 3 demos in the source code repository:

  1. body_pose: Detect 18-point posture of the human body, only support image recognition
  2. image_classify: Object recognition classification, only supports image recognition
  3. person_detect: Human body detection, supports image recognition and camera recognition

There are compilation scripts, makefiles and source codes in each directory. Take person_detect as an example.

$ cd workspace/aml_npu_nnsdk_app/person_detect_640x384
$ ls
build-cv3.sh  include  makefile-cv3.linux  person_detect_640x384_camera.cpp  person_detect_640x384_picture.cpp  README.cn.md  README.md
  1. build-cv3.sh : Compiled script
  2. makefile-cv3.linux : Compilesd makefile
  3. person_detect_640x384_camera.cpp: Source code for image recognition
  4. person_detect_640x384_picture.cpp: Source code for camera recognition

Compilation method

Please refer to get SDK #Get-SDK

Here also take person_detect as an example,

$ cd workspace/aml_npu_nnsdk_app/person_detect_640x384
$ ./build-cv3.sh /path/to/aml_npu_sdk/linux_sdk/linux_sdk
  COMPILE /home/yan/data/git/npu/aml_npu_nnsdk_app/person_detect_640x384/person_detect_640x384_picture.cpp
  COMPILE /home/yan/data/git/npu/aml_npu_nnsdk_app/person_detect_640x384/person_detect_640x384_camera.cpp

Compilation will generate the generated file in cv3_output,

$ cd workspace/aml_npu_nnsdk_app/person_detect_640x384/cv3_output
$ ls
person_detect_640x384_camera  person_detect_640x384_camera.o  person_detect_640x384_picture  person_detect_640x384_picture.o

Among them, person_detect_640x384_camera and person_detect_640x384_picture are the generated executable files

How to Run

Here also take person_detect as an example,

  1. Obtain the nb file https://github.com/khadas/AML_NN_SDK, the nb file corresponding to person_detect is:
$ mkdir board_space
$ cd board_space
$ wget https://github.com/khadas/AML_NN_SDK/raw/master/Model/DDK6.4.4.3/88/person_detect_88.nb  [VIM3]
$ wget https://github.com/khadas/AML_NN_SDK/raw/master/Model/DDK6.4.4.3/99/person_detect_99.nb  [VIM3L]

  1. Copy the executable file compiled on the PC to the board

  2. Run

Identify the picture

$ /path/to/person_detect_640x384_picture board_space/person_detect_88.nb /path/to/picture  [VIM3] 
$ /path/to/person_detect_640x384_picture board_space/person_detect_99.nb /path/to/picture  [VIM3L]

Recognition camera

$ /path/to/person_detect_640x384_camera board_space/person_detect_88.nb /dev/videoX   [VIM3]
$ /path/to/person_detect_640x384_camera board_space/person_detect_99.nb /dev/videoX   [VIM3L]

Note :

Just a simple template repository, please refer to the documentation for detailed API introduction.