Installing TensorRT on Ubuntu20.04

-

1. Install CUDA, cuDNN

CUDA:
(1) Remove old version
sudo apt-get purge nvidia*
sudo apt-get autoremove
sudo apt-get autoclean
sudo rm -rf /usr/local/cuda*

(2) Download and install CUDA toolkit 11.3 [1]
$ wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
$ sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
$ wget https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda-repo-ubuntu2004-11-3-local_11.3.0-465.19.01-1_amd64.deb
$ sudo dpkg -i cuda-repo-ubuntu2004-11-3-local_11.3.0-465.19.01-1_amd64.deb
$ sudo apt-key add /var/cuda-repo-ubuntu2004-11-3-local/7fa2af80.pub
$ sudo apt-get update
$ sudo apt-get -y install cuda

(3) Check
$ cd /usr/local/cuda-11.3/samples/1_Utilities/deviceQuery
$ sudo make
$ ./deviceQuery

ex:

cuDNN:
(1) Download cuDNN 8.2.1 [2] 

(2) Install
extract cudnn-11.3-linux-x64-v8.2.1.32.tgz
$ cd cuda
$ sudo cp cuda/include/cudnn.h /usr/local/cuda-11.3/include/ 
$ sudo cp cuda/include/cudnn_version.h /usr/local/cuda-11.3/include/
$ sudo cp cuda/lib64/libcudnn* /usr/local/cuda-11.3/lib64/ 
$ sudo chmod a+r /usr/local/cuda-11.3/include/cudnn.h 

(3) Check
$ cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2

2. Download the corresponding .deb file for installing TensorRT [3]

* EA version stands for early access (It is before actual release). GA stands for general availability. GA is stable version and completely tested.

$ sudo dpkg -i nv-tensorrt-repo-ubuntu2004-cuda11.3-trt8.0.1.6-ga-20210626_1-1_amd64.deb
$ sudo apt-key add /var/nv-tensorrt-repo-ubuntu2004-cuda11.3-trt8.0.1.6-ga-20210626/7fa2af80.pub
$ sudo apt-get update
$ sudo apt-get install tensorrt
$ sudo apt-get install python3-libnvinfer-dev

Check: 

$ dpkg -l | grep TensorRT


* Remember to install pycuda for using tensorrt in Python
$ pip3 install 'pycuda<2021.1'
or 
Download the source file [4]
$ ./configure.py
$ make
$ sudo make install

* Useful command for checking Ubuntu, CUDA, cuDNN version

Ubuntu:

$ lsb_release -a

CUDA:

$ nvcc -V

or

$ nvidia-smi 

or

$ import torch

$ print(torch.version.cuda)

cuDNN:

cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

or 

$ import torch

$ print(torch.backends.cudnn.version())


Reference:

[1] CUDA Toolkit 11.3 Downloads

[2] cuDNN Library for Linux (x86_64)

[3] https://developer.nvidia.com/nvidia-tensorrt-8x-download

[4] https://pypi.org/project/pycuda/#modal-close

[5] https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html

[6] https://zhuanlan.zhihu.com/p/392143346

[7] https://www.uj5u.com/qita/299907.html

留言

張貼留言

這個網誌中的熱門文章

Useful PX4 Parameters

-
1. For setting companion link SYS_COMPANION (Before ver. 1.9.0) MAV_0_CONFIG, MAV_0_MODE (After ver. 1.9.0) 2. EKF2 configuration EKF2_AID_MASK EKF2_HGT_MODE EKF2_EV_DELAY (Reduce this parameter if high rate external vision is available) EKF2_EV_POS_X, EKF2_EV_POS_Y, EKF2_EV_POS_Z 3. Offboard mode parameters ( The vehicle will exit the mode if target setpoints are not received at a rate of > 2Hz ) COM_OBL_RC_ACT (Default: position mode, I set land mode) COM_OBL_ACT COM_OF_LOSS_T 4. Position loss failsafe COM_POSCTL_NAVL 5. Land mode MPC_LAND_SPEED COM_DISARM_LAND: Time-out for auto disarm after landing. (Default: 0, I set 3.) 6. TBD Reference: [1]  https://dev.px4.io/en/advanced/parameter_reference.html [2]  https://dev.px4.io/en/ros/external_position_estimation.html#tuning-EKF2_EV_DELAY [3]  https://docs.px4.io/en/config/safety.html#qgc_safety_setup
閱讀完整內容

Tuing PID parameters in QGroundcontrol (2)

-
圖片
Based on tips for tuning PX4 PID parameters written in my previous post [1] or even using the default values directly, you will be able to fly your vehicle freely in manual or stabilized flight mode.   However, if you are using an unique vehicle (usually assembly by yourself and only equipped with sensors you desire) and struggling for finding the best PID parameters for your applications. I think following 8 tips will help you to get ideal PID parameters.  This is the vehicle I made based on LHI QAV250 fiber frame, with TX2, high-speed camera, etc.[2]  Tip1:  Tuning order: Rate controller -> attitude controller -> position controller The PX4 controllers are layered [3], which means a higher-level controller passes its results to a lower-level controller. Thus the PID tuning needs to be done from lowest-level controller to highest-level controller. Tip2:  Check if  PWM_MIN  is set correctly. Remove propellers Arm the vehic...
閱讀完整內容

Matlab Pixhawk Support Package installation (Windows)

-
圖片
1. Download the pixhawk support package [1] 2. Open MATLAB R2017b and navigate to the file location. 3. Click "PX4PSP_v3_0_4_351_R2017b" and install it 4. Create a folder for firmware download path (ex: D/PX4) 5. Setting bash environment for windows 10 [2]  (1) Active "Windows subsystem for linux" (2) Download and install Ubuntu (3) Using the terminal like in Ubuntu environment ex: 6. Install all necessary toolchain with " windows_bash_nuttx.sh " [3] or using Cygwin Toolchian [4] or [5] (python27 would show in following folder) 7. Key in command " PixhawkPSP(' D:\PX4 ') " in the Matlab 8. Make sure windows 10 bash and python path are correct => Validate paths 9. Download firmware to the folder we specify before and then validate firmware. 10. Selecting cmake configuration [6] 11. Build firmware *Error: make command not found   Solution: [7] T...
閱讀完整內容