How To Use Pre-Built NPU Demo

Pre-Build NPU demo Repository

aml_npu_demo_binaries repository are executable files repositories, this repository are used to run NPU demo on khadas VIM3/VIM3L

Repository information

  • Repository address: gitlab.com/khadas/aml_npu_demo_binaries

  • Repository structure

    1. detect_demo/detect_demo_khadas : These two demos are demos that use cameras to recognize objects for classification
    2. detect_demo_picture : This demo is to identify and classify multi-object pictures
    3. inceptionv3 : This demo is a demo that recognizes pictures and makes top5 predictions

How To Use

preparation

  1. Please run ubuntu firmware on VIM3 and upgrade the system through sudo apt update && sudo apt upgrade

  2. detect_demo_picture/detect_demo/detect_demo_khadas need to be used in framebuffer mode, use Ctrl+Alt+F1 to switch to framebuffer mode

  3. Clone the repository to VIM3/VIM3L

$ cd ${workspace}
$ git clone https://gitlab.com/khadas/aml_npu_demo_binaries

Use detect_demo or detect_demo_khadas or detect_demo_picture

Install demo:

$ cd ${workspace}/aml_npu_demo_binaries
$ sudo ./INSTALL

After installation, the instructions to install the README file can run the demo.

detect_demo_uvc is the running file of the USB camera
detect_demo_mipi是mipi摄像头的运行文件

uninstall demo:

$ sudo ./UNINSTALL

Use inceptionv3

If you board is VIM3

$ cd ${workspace}/aml_npu_demo_binaries/inceptionv3/vim3

If you board is VIM3L

$ cd ${workspace}/aml_npu_demo_binaries/inceptionv3/vim3l

run demo

$ ./inceptionv3 inception_v3.nb /path/to/picture

Image format needs to use jpg format

aml_npu_app Repository

aml_npu_app repository is used to compile demo repository, this repository runs on PC

Repository information

  • Repository address: gitlab.com/khadas/aml_npu_app

  • Repository structure

    1. DDK_6.3.2/DDK_6.3.2.3/DDK_6.3.2.5/DDK_6.3.3.4 is the source code of different versions of library files
    2. detect_library/model_code point to the latest version of the library source code
    3. detect_library/source_code generate detect.so
    4. detect_library/yolo_demo_gst_uvc or detect_library/yolo_demo_mipi generate USB demo or MIPI demo
    5. detect_library/sample_demo Generate a demo that recognizes the picture.

How To Use

Preparation

  1. Apply sdk,and unzip.

  2. clone repository to you PC

$ cd ${workspace}
$ git clone https://gitlab.com/khadas/aml_npu_app

Compile

Enter the location where you need to compile, here takes yolov3 as an example.

$ cd ${workspace}
$ cd aml_npu_app/detect_library/model_code/detect_yolo_v3
$ ./build_vx.sh /path/to/sdk/linux_sdk/linux_sdk

After compilation, the files in the bin_r directory are the files generated by the compilation

$ ls bin_r/
libnn_yolo_v3.so vnn_yolov3.o yolo_v3.o yolov3_process.o