Installation
1. Create a Python environment
Anaconda is recommended for creating the Python environment. XenseSDK currently supports Python 3.9 through 3.12; the example below uses Python 3.9.19 to match the project examples:
conda create -n xenseenv python=3.9.19
conda activate xenseenv
2. Install CUDA and cuDNN
CUDA Toolkit and cuDNN are not installed automatically with XenseSDK and must be installed and configured separately for your environment. For GPU inference, their versions must be compatible with onnxruntime-gpu. The verified reference environment uses CUDA 11.8 + cuDNN 8.9.2.26; for other environments, follow the release notes and GPU driver compatibility.
conda search cudnn
conda search cudatoolkit
conda install cudnn==8.9.2.26 cudatoolkit==11.8.0
On Linux, if the dynamic libraries are not found at runtime, you can temporarily set:
export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$CONDA_PREFIX/lib64:$LD_LIBRARY_PATH"
Without an NVIDIA GPU, you can first disable inference and verify the communication and data paths by passing disable_infer=True to Sensor.create. Use infer_mode or overrides when you need to switch inference size or mode.
3. Install the package
The recommended default is the full installation, which includes the core features, local ONNX inference, and visualization:
pip install 'xensesdk[full]'
When installing from a local wheel, choose the file matching your Python, OS, and architecture:
pip install 'xensesdk-x.y.z-cp39-cp39-win_amd64.whl[full]'
The current package provides the following installation options:
| Installation command | Installed components | When to use it |
|---|---|---|
pip install xensesdk | Core dependencies only | When local ONNX inference and built-in visualization are not required |
pip install 'xensesdk[onnx]' | Core dependencies + onnxruntime-gpu | When local ONNX inference is required but built-in visualization is not |
pip install 'xensesdk[viz]' | Core dependencies + OpenGL, Assimp, and ImGui visualization dependencies | When built-in visualization is required but local ONNX inference is not |
pip install 'xensesdk[full]' | Core dependencies + ONNX inference + visualization dependencies | Recommended full installation |
Run python -c "from xensesdk import Sensor; print(Sensor.scanSerialNumber())". If the result is empty, check USB connectivity, device permissions, compute-unit networking, and sensor configuration files.