Jetson-inference environment construction

The Raspberry Pi motherboard series does not currently support this tutorial.

一、Install jetson-inference environment

1.Instructions before use

This tutorial is suitable for building a jetson nano image independently. If you use the YAHBOOM version of the image directly, you can ignore this tutorial.

2.The environment version configuration of this tutorial is as shown in the figure：

If you don’t want to build it completely by yourself, you can use the jetson-inference compressed package we provide, pass the compressed package into jetson nano, unzip it, and start directly from "Installing Modules"

3. Start building

3.1 Download required dependencies

sudo apt-get update
sudo apt-get install git cmake

3.2 Download related source code

xxxxxxxxxx
git clone https://github.com/dusty-nv/jetson-inference
cd jetson-inference
git submodule update --init 

3.3 Download the relevant python module

Find the file torch-1.8.0-cp36-cp36m-linux_aarch64.whl from the attachment built by our environment and transfer it to jetson nano

xxxxxxxxxx
sudo apt-get install libpython3-dev python3-numpy
sudo apt-get install python3-scipy
sudo apt-get install python3-pandas
sudo apt-get install python3-matplotlib
sudo apt-get install python3-sklearn
pip3 install torch-1.8.0-cp36-cp36m-linux_aarch64.whl

3.4 Make changes to files

Edit jetson-inference/CMakePrebuild.sh. Comment out ./download-models.sh (add a # comment in front) as shown in the figure)

4.Install model

Method 1: You can perform the following steps

xxxxxxxxxx
cd jetson-inference/tools
./download-models.sh

After making a selection, the model will be automatically downloaded to the file path of data/network.

Method 2: You can find the packages required by jetson-inference in the attachments we provide for environment construction, transfer the compressed packages inside to jetson-inference/data/network of jetso nano, and then decompress them. Unzip command

xxxxxxxxxx
for tar in *.tar.gz; do tar xvf $tar; done

Note:

1. For decompressing multiple .gz files, use this command: for gz in *.gz; do gunzip $gz; done

2. For decompressing multiple .tar.gz files, use the following command for tar in *.tar.gz; do tar xvf $tar; done

5. Start compiling

xxxxxxxxxx
cd jetson-inference    
mkdir build   
cd build
cmake ../
make (or make -j4)    #（In the build directory）
sudo make install      #（In the build directory）

If an error is reported during the process, it means that the source code download is incomplete. please go back to step 3.2 and execute the command git submodule update --init.

6.Verify whether the installation is successful

xxxxxxxxxx
cd jetson-inference/build/aarch64/bin

./imagenet-console ./images/bird_0.jpg output.jpg
# After the execution waits for a long time, the following appears (the first time it takes a long time, the subsequent execution will be very fast)

Find the corresponding directory and view output.jpg as shown below. The recognition result will be displayed at the top of the picture.

二、Install Mediapipe environment

1.Preparing Files

Transfer the two files bazel and mediapipe-0.8-cp36-cp36m-linux_aarch64.whl in the attachment of the environment setup to the jetson nano

2.Install bazel

Open a terminal and run the following command

xxxxxxxxxx
sudo chmod +x bazel
mv bazel /usr/local/bin

Check whether the installation of bazel is complete. If the version number can be printed, the installation is complete.

xxxxxxxxxx
bazel --version

3.Install mediapipe

Open a terminal and run the following command

xxxxxxxxxx
pip3 install opencv-contrib-python==3.4.18.65
pip3 install mediapipe-0.8-cp36-cp36m-linux_aarch64.whl
pip3 uninstall opencv-contrib-python

Verify successful installation

xxxxxxxxxx
python3

import mediapipe as mp

appendix

Other reference tutorial URLs： 1.https://blog.csdn.net/aal779/article/details/122055432

2.https://github.com/dusty-nv/jetson-inference/blob/master/docs/building-repo-2.md

3.https://blog.csdn.net/weixin_43659725/article/details/120211312