Building Python Bindings for Conda Virtual Environment on Jetson

[happydrones]

Hi everyone,

I'm working on building Librealsense (v2.56.3) on a Jetson agx orin device with JetPack6.2 [L4T 36.4.3]
and have been following the instructions to compile from source. The standard procedure involves cloning the repository and running CMake with the following commands:

# Clone the repository
git clone https://github.com/IntelRealSense/librealsense.git

# Navigate to the repository and create a build directory
cd librealsense
mkdir build
cd build

# Run CMake with specified options for Python bindings
cmake .. -DBUILD_EXAMPLES=true -DBUILD_PYTHON_BINDINGS=bool:true -DPYTHON_EXECUTABLE=/usr/bin/python3.10 -DBUILD_DOCUMENTATION=false -DCHECK_FOR_UPDATES=false

# Build the project
make

After building, I was able to successfully import pyrealsense2 using the system Python (i.e. when I run python after deactivating conda). However, I recently created a new conda virtual environment called yolov5:

conda create -n yolov5 python=3.12.9
conda activate yolov5

In this conda environment, when I try to run python and do import pyrealsense2, it fails.error below:

(yolov5) oscar@ubuntu:~/miniconda3/envs/yolov5/bin$ python
Python 3.13.2 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 19:03:27) [GCC 11.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyrealsense2
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    import pyrealsense2
ModuleNotFoundError: No module named 'pyrealsense2'

I suspect that changing the CMake parameter from:-DPYTHON_EXECUTABLE=/usr/bin/python3.10
to:-DPYTHON_EXECUTABLE=/home/oscar/miniconda3/envs/yolov5/bin/python
might help. However, I am not sure if any other CMake parameters like -DBUILD_EXAMPLES=true -DBUILD_PYTHON_BINDINGS=bool:true -DPYTHON_EXECUTABLE=/usr/bin/python3.10 -DBUILD_DOCUMENTATION=false -DCHECK_FOR_UPDATES=false need to be adjusted in order to properly compile the Python bindings so that they work with the conda virtual environment.

My Questions:

1. Has anyone successfully built Librealsense’s Python bindings for a conda environment on Jetson?

2. Besides changing the PYTHON_EXECUTABLE path, are there any additional CMake options or modifications that need to be made to support the conda virtual environment?

3. Any additional tips for ensuring that the built pyrealsense2 is correctly linked against the libraries within the conda environment?

Any help or suggestions would be greatly appreciated!

Thanks in advance.

[MartyG-RealSense]

Hi @happydrones Please first try changing your pyrealsense2 import line to the one below.

import pyrealsense2.pyrealsense2 as rs

Does the ModuleNotFoundError still occur when using that modified command?

I also recommend using the command which python3 to check whether the PYTHONPATH is the same when using Conda, as for some Conda users they have a different path than when using Python 3. After finding the path, input the PYTHONPATH command into the terminal before running your script. Use the command below but change the path to whatever 'which python3' provides.

export PYTHONPATH=$PYTHONPATH:/usr/local/lib

---

I did not find a previous successful example of using Conda with librealense and pyrealsense2 on Jetson. There is a librealsense / pyrealsense2 guide for Conda at the link below though that is not specific to Jetson.

https://github.com/conda-forge/librealsense-feedstock

[happydrones]

hi @MartyG-RealSense

Hi @happydrones Please first try changing your pyrealsense2 import line to the one below.
import pyrealsense2.pyrealsense2 as rs
Does the ModuleNotFoundError still occur when using that modified command?
I have tried this few times errors below:

(base) oscar@ubuntu:~$ conda deactivate
oscar@ubuntu:~$ python
Python 3.10.12 (main, Feb  4 2025, 14:57:36) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyrealsense2
>>> exit()
oscar@ubuntu:~$ conda activate
(base) oscar@ubuntu:~$ which python
/home/oscar/miniconda3/bin/python
(base) oscar@ubuntu:~$ which python3
/home/oscar/miniconda3/bin/python3
(base) oscar@ubuntu:~$ export PYTHON=$PYTHON:/home/oscar/miniconda3/bin/python
(base) oscar@ubuntu:~$ export PYTHON=$PYTHON:/home/oscar/miniconda3/bin/python3
(base) oscar@ubuntu:~$ python
Python 3.12.9 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 8:48:18) [GCC 11.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyrealsense2.pyrealsense2
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'pyrealsense2'
>>> import pyrealsense2
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'pyrealsense2'
>>> exit()
(base) oscar@ubuntu:~$ 

My idea is whether I can download the source code from GitHub and then change the CMake command from -DPYTHON_EXECUTABLE=/usr/bin/python3.10
to -PYTHON_EXECUTABLE=/home/oscar/miniconda3/envs/[virtual_env_name]/bin/python. This is because I want to create a Python wrapper specifically for the Python interpreter in the conda virtual environment. However, I am unsure whether this will affect the system's default Python. Alternatively, I might not actually be concerned about whether it affects the system's Python.

[MartyG-RealSense]

You should not need to use the -DPYTHON_EXECUTABLE flag unless you have more than one version of Python (such as Python 2 and Python 3) installed on your computer, as the flag tells the installation which Python version to use with the pyrealsense2 wrapper.