Application of ncnn neural network computing framework in orange school orangepi 3 lts development board

 

1、 tencent ncnn The source code download command is as follows  
1） The first method ： download Orang Pi Baidu cloud disk provides ncnn.tar.gz Compressed package  
a) You can download it from the baidu cloud disk link below ncnn.tar.gz Source compression package . Get into ncnn writing You can see it in the clip  

b) Download the ncnn.tar.gz After compressing the package , First of all, will ncnn.tar.gz Upload to development board linux system Middle series  
c) Then use the following command to extract ncnn.tar.gz  

2） The second method ： Use git Command to download the source code directly , But if it doesn't solve the problem of development board access github The problem of , It is difficult to download successfully . If you visit github No problem , It is recommended to use this Methods , Because this way can ensure that the code is up-to-date . 

 

2、 Then install the dependent package

3、 Then start compiling ,ncnn The compile command looks like this  

explain ： Without any heat dissipation measures , Compile directly on the development board ncnn About need 15 minute , Please wait patiently for the compilation to complete . If a fan is added to cool the development board , The speed should be faster .

4、ncnn There are some test examples in , such as squeezenet The test commands and results are as follows  

 

5、benchncnn It can be used to test the reasoning performance of Neural Networks , The test method is as follows  
1） Compile generated benchncnn The executable file is in the following path , Pay attention to the following command execution path The diameter is ncnn Top level directory of source code

2） First, you need to benchncnn Copied to the benchmark Directory  

3） benchncnn The usage of is as follows  

4） benchncnn Use cpu The test results are shown below  

a) Debian Bullseye Linux5.16 Server version system test results  

6、NanoDet It's an ultra fast and lightweight mobile terminal Anchor-free Target detection model . Tested by The method is as follows  
1） Compile generated nanodet The executable file is in the following path , Note that the following command execution path is ncnn A directory on the source code  

2） First, create a new one nanodet_demo Folder  

3） The generated... Will then be compiled nanodet Copy the executable program to nanodet_demo In the folder  

4） Then you need to download nanodet Model file and upload to nanodet_demo In the folder  
a) nanodet The download address of the model file is as follows   

b) Open the link above , look for nanodet_m.bin and nanodet_m.param These two documents , And download it , Then upload to the development board Linux Systematic nanodet_demo In the folder   

c) here nanodet_demo There should be the following three files in the folder  

5） Then you need to put the image you want to detect in nanodet_demo In the folder , For example, the following one has Many pictures of cars ( You can use your mobile phone to take a few pictures of traffic or animals ) 

6） Then run the following command to use nanodet Target detection ,car.jpg Please replace with The name of your picture

7） The results of the test will be saved in the file named image.png In the picture of  

8） If you use a desktop version Linux System , You can open it directly image.png see , If so It uses the server version Linux System , Can be image.png Copy to the computer to view , image.png The content is shown in the figure below , You can see that the top left corner of the identified object will show the kind of object Percentage of class and confidence  

7、 For testing purposes benchncnn and nanodet, I sorted out a document that only contains benchncnn and nanodet The executable file and the model file required for the test are packaged into a ncnn_test_demo.tar.gz Put the compressed package On Baidu cloud disk , No need to download and compile ncnn Source code , With this executable program, you can directly Then the test begins

1） You can download it from the baidu cloud disk link below ncnn_test_demo.tar.gz Compressed package . Get into ncnn You can see in the folder  

2） Download the ncnn_test_demo.tar.gz After compressing the package , First of all, will ncnn_test_demo.tar.gz Pressure Shrink the package and upload it to the development board linux In the system

3） Then use the following command to extract ncnn_test_demo.tar.gz 

4） Decompress and enter ncnn_test_demo You can see that the directory contains benchncnn_demo and nanodet_demo Two subfolders , They are used to test benchncnn and nanodet 

5） Get into benchncnn_demo Folder , And then run ./benchncnn 4 $(nproc) 0 -1 This life So we can directly test the reasoning performance of neural network  

6） Get into nanodet_demo Folder , And then run ./nanodet car.jpg This order can be straight Connect to use nanodet To detect car.jpg The object in the picture , You can also put the image you want to detect Put the film on nanodet_demo In the folder , And then use nanodet To detect