NPU SDK is a collection of tools for converting AI models and compiling aml_npu_sdk

Get SDK

The SDK needs to be applied for by email, and an email will be sent to your mailbox after the application. application address

SDK directory structure description

Enter the SDK directory,

$ cd {workspace}/aml_npu_sdk
$ ls
acuity-toolkit  android_sdk  Dockerfile  docs  LICENSE  linux_sdk  README.md  toolchains

The SDK is mainly divided into several SDK, conversion tools and compilation tools, and docs.

acuity-toolkit    #Conversion tool , used to convert AI models
android_sdk       #Android SDK 
linux_sdk         #linux SDK ,use for compiling `aml_npu_app`
docs              #Conversion related documents collection
toolchains        #Compile toolchain directory

Docs description

entre Docs directory,

$ cd {workspace}/aml_npu_sdk/docs/en
$ ls
'AMLNN Convolution Acceleration Tips.pdf'                  'Model_Transcoding and Running User Guide_V0.6.pdf'                              'NN Tool FAQ (0.1).pdf'
'Android&Linux_Compilation and Integration Guide_0.2.pdf'  'Neural Network Layer and Operation Support Guide (01)(ref.v1.13-20200323).pdf'

The document records a series of processes from conversion to integration, as well as some common problems

1. 'Android&Linux_Compilation and Integration Guide_0.2.pdf'                         #Android&&linux compilation and integration guide, mainly explaining how to use the converted code
2. 'NN Tool FAQ (0.1).pdf'                                                           #Conversion tool FAQ document, which records common conversion problems
3. 'Model_Transcoding and Running User Guide_V0.6.pdf'                               #Model conversion document, a detailed description of how to convert
4. 'AMLNN Convolution Acceleration Tips.pdf                                          #AMLNN convolution acceleration docs
5. 'Neural Network Layer and Operation Support Guide (01)(ref.v1.13-20200323).pdf'   #Supported network layer and operator documentation

Conversion tool description

acuity-toolkit is the conversion tool directory,

$ cd {workspace}/aml_npu_sdk/acuity-toolkit
$ ls
bin  conversion_scripts  ReadMe.txt  requirements.txt

The main directory of interest is conversion_scripts

1. bin                   #Conversion is a collection of various tools used, most of which are not open source.
2. conversion_scripts    #Conversion script directory, convert AI model location
3. ReadMe.txt            #ReadMe.txt file explains how to convert and use
4. requirements.txt      #Conversion tool dependent environment

Dependent installation

The environment dependency package required by the conversion tool can be installed directly on the PC or installed through the virtual environment virtualenv

$ cd {workspace}/aml_npu_sdk/acuity-toolkit
$ cat requirements.txt
tensorflow==2.0.0
astor==0.8.0
numpy==1.18.0
scipy==1.1.0
Pillow==5.3.0
protobuf==3.11.2
networkx>=1.11
image==1.5.5
lmdb==0.93
onnx==1.6.0
h5py==2.10.0
flatbuffers==1.10
matplotlib==2.1.0
dill==0.2.8.2
ruamel.yaml==0.15.81
ply==3.11
torch==1.2.0

Among them, tensorflow==2.0.0 can be replaced by tensorflow==2.0.0a0

Conversion script usage

The conversion script is in the conversion_scripts directory,

$ cd {workspace}/aml_npu_sdk/acuity-toolkit/conversion_scripts
$ ls
0_import_model.sh  1_quantize_model.sh  2_export_case_code.sh  data  dataset.txt  extractoutput.py  inference.sh  mobilenet_tf.data  mobilenet_tf.json  mobilenet_tf.quantize  model  normal_case_demo

Use scripts to convert AI models

$ cd {workspace}/aml_npu_sdk/acuity-toolkit/conversion_scripts
$ bash 0_import_model.sh && bash 1_quantize_model.sh && bash 2_export_case_code.sh

After the conversion is completed, you can see the converted code in the nbg_unify_xxxx directory, here is the built-in model as an example

$ cd {workspace}/aml_npu_sdk/acuity-toolkit/conversion_scripts/nbg_unify_mobilenet_tf
$ ls
BUILD   makefile.linux   mobilenettf.vcxproj  vnn_global.h       vnn_mobilenettf.h   vnn_post_process.h  vnn_pre_process.h
main.c  mobilenet_tf.nb  nbg_meta.json        vnn_mobilenettf.c  vnn_post_process.c  vnn_pre_process.c

For the setting of conversion parameters, please refer to’Model Conversion Operation User Guide (0.6).pdf’ in Docs

linux SDK instructions

Enter the linux SDK directory

$ cd {workspace}/aml_npu_sdk/linux_sdk
$ ls
demo  linux_sdk  linux_sdk_6.4.0.10  linux_sdk_6.4.2.1  linux_sdk_6.4.3

Here you can see a simple demo that has been converted, and various versions of the linux SDK.

1. demo                     #The executable file and source code of the converted inception model
2. linux_sdk                #Link to the latest SDK
3. linux_sdk_x.x.x.x        #Different versions of linux SDK

Enter linux_sdk, you can see the main components of the sdk

$ cd {workspace}/aml_npu_sdk/linux_sdk/linux_sdk
$ ls 
acuity-ovxlib-dev  build  common.target  linux_build_sample.log  makefile.linux.def

Mainly used parts:

1. `acuity-ovxlib-dev`          #Mainly placed the `ovxlib` library needed for compilation
2. `build/sdk/drivers_xx`       #Mainly placed the system libraries needed for compilation
3. `build/sdk/opencvX`          #Mainly the opencv library used during compilation