Instructions for installing opencv with CUDA support for Jetson Orin Nano Jetpack 6.2

[TegwynTwmffat]

Anybody got up to date instructions for installing opencv with cuda support? I tried the old version from jetsonhacks video, but it’s now out of date. Tried various other permutations, but to no avail.

nvidia-l4t-core 36.4.4-20250616085344
Cuda v.12.6

Thanks !!!

[carolyuu]

*** Please note that this reply is generated by LLM automatically ***

Installing OpenCV with CUDA support on Jetson Orin Nano can be a bit tricky, but I’ll provide you with the most up-to-date instructions.

Prerequisites:

1. Make sure you have the latest JetPack version installed on your Jetson Orin Nano. You can check the version by running dpkg -l | grep jetpack in the terminal.
2. Ensure that CUDA is installed and configured correctly. You can verify this by running nvcc --version in the terminal.

Installing OpenCV with CUDA support:

1. Method 1: Using the pre-built OpenCV package with CUDA support

You can install the pre-built OpenCV package with CUDA support using the following command:

sudo apt-get update
sudo apt-get install libopencv-dev=4.5.1-1+cuda11.1

This will install OpenCV 4.5.1 with CUDA 11.1 support.

2. Method 2: Building OpenCV from source with CUDA support

If you prefer to build OpenCV from source, you can follow these steps:

a. Install the required dependencies:

sudo apt-get update
sudo apt-get install build-essential cmake git libgtk-3-dev libavcodec-dev libavformat-dev libswscale-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev

b. Clone the OpenCV repository:

git clone https://github.com/opencv/opencv.git
cd opencv

c. Create a build directory and navigate to it:

mkdir build
cd build

d. Configure the build with CUDA support:

cmake -D CMAKE_BUILD_TYPE=Release -D WITH_CUDA=ON -D CUDA_ARCH_BIN=8.6 -D CUDA_ARCH_PTX=8.6 ..

Note: The CUDA_ARCH_BIN and CUDA_ARCH_PTX flags are set to 8.6, which is the architecture of the Jetson Orin Nano.

e. Build and install OpenCV:

make -j4
sudo make install

This will build and install OpenCV with CUDA support.

Verification:

After installation, you can verify that OpenCV is working with CUDA support by running the following command:

python -c "import cv2; print(cv2.getBuildInformation())"

This should display the OpenCV build information, including the CUDA version.

Comparison with the old method:

The old method from the JetsonHacks video is outdated and may not work with the latest JetPack version. The new method described above uses the pre-built OpenCV package with CUDA support or builds OpenCV from source with CUDA support, which is more reliable and efficient.

Additional tips:

• Make sure to update your JetPack version regularly to ensure you have the latest security patches and features.
• If you encounter any issues during the installation process, you can try cleaning up the build directory and starting from scratch.
• You can also try building OpenCV with CUDA support using the cmake command with the -D WITH_CUDA=ON flag, but this may require additional dependencies and configuration.

*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***

[TegwynTwmffat]

No workio.
Please dont reply with LLM again.

What I have been working on is variations of the below:

source python_env_02/bin/activate
sudo apt install -y build-essential cmake git pkg-config
sudo apt install -y python3-dev python3-numpy python3-pip
pip install numpy scipy matplotlib pandas

mkdir -p ~/opencv_build && cd ~/opencv_build
git clone --depth 1 --branch 4.7.0 GitHub - opencv/opencv: Open Source Computer Vision Library
git clone --depth 1 --branch 4.7.0 GitHub - opencv/opencv_contrib: Repository for OpenCV's extra modules
cd ~/opencv_build/opencv
mkdir build && cd build

cmake
-D CMAKE_BUILD_TYPE=RELEASE
-D CMAKE_INSTALL_PREFIX=/usr/local
-D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules
-D WITH_CUDA=ON
-D WITH_CUDNN=ON
-D ENABLE_FAST_MATH=1
-D CUDA_FAST_MATH=1
-D WITH_CUBLAS=1
-D WITH_GSTREAMER=ON
-D WITH_TBB=ON
-D BUILD_NEW_PYTHON_SUPPORT=ON
-D BUILD_opencv_python3=ON
-D OPENCV_GENERATE_PKGCONFIG=ON
-D OPENCV_ENABLE_NONFREE=ON
-D PYTHON3_EXECUTABLE=~/python_env_02/bin/python
-D PYTHON3_INCLUDE_DIR=/usr/include/python3.10
-D PYTHON3_LIBRARY=/usr/lib/aarch64-linux-gnu/libpython3.10.so
-D PYTHON3_PACKAGES_PATH=~/python_env_02/lib/python3.10/site-packages
-D CUDA_ARCH_BIN=8.7
-D INSTALL_PYTHON_EXAMPLES=OFF
-D INSTALL_C_EXAMPLES=OFF
-D BUILD_EXAMPLES=OFF ..

make -j6
sudo make install

cd ~/python_env_02/lib/python3.10/site-packages
ln -s /usr/local/lib/python3.10/site-packages/cv2.cpython-310-aarch64-linux-gnu.so cv2.so

[DaneLLL]

Hi,
If there is no at-hand deb package, you can manually build it from source code. Please refer to Jetson AGX Orin FAQ and try the script.

[TegwynTwmffat]

That worked ok in one of my python environments when running with ultralytics yolo, but in another i got the following error:
Traceback (most recent call last):

File “/home/nano/Documents/yolo_stuff/test_yolo_on_video_file_onions_16.py”, line 21, in

from ultralytics import YOLO

File “/home/nano/python_env_01/lib/python3.10/site-packages/ultralytics/init.py”, line 12, in

from ultralytics.utils import ASSETS, SETTINGS

File “/home/nano/python_env_01/lib/python3.10/site-packages/ultralytics/utils/init.py”, line 25, in

import torch

File “/home/nano/python_env_01/lib/python3.10/site-packages/torch/init.py”, line 361, in

from torch._C import *  # noqa: F403

ImportError: libcusparseLt.so.0: cannot open shared object file: No such file or directory

… searching for the file yielded no results in that environment, but i found it installed in another one at:
/home/nano/python_env_01/lib/python3.10/site-packages/nvidia/cusparselt/lib/libcusparseLt.so.0

… not sure how it got there, maybe while installing jetson-inference?

[lenamphong]

i also had the same problem installing ultralytics on orin nano.
For a newbie like me it was horrible.