This article provides backgroud, download links and installation instructions for the stand-alone GPU package installation.
TLDR
- If you use ML and have a compatible GPU you should install this and use it
- The installation is normally part of the standard installation process
- If you've already installed the software, or you're installing it on an offline workstation, you can download and install this package independently using the instructions below
Introduction
With the release of Vision4D 3.6, arivis added the ability to use Deep Learning (DL) inference for segmentation of images. This can significantly improve the accuracy of segmentation of complex structures in a range of applications and images.
To optimize the DL operations we use the ONNX runtime libraries. These can be used with both GPU and CPU, but using the GPU to run these operations can dramatically improve the speed of processing. However, the GPU libraries are quite large (600MB) and only compatible with NVIDIA GPUs. Since not all users will require this, either because they have no need for DL or no compatible GPU, we do not include the GPU package as part of the Vision4D installation files and instead download it on request during the installation process.
Installing the GPU Package for arivis Vision4D
When installing Vision4D, the user is prompted for a variety of preferences, including whether to install the GPU package:
If the computer where the installation takes place is connected to the internet, selecting this option will start the download and installation of the additional components. Users can simply proceed with the installation as normal with no additional action.
However, if the computer in question is not connected to the internet, this will not be possible. Instead, the GPU package will need to be downloaded and installed independently. The GPU package is available to download here. Once downloaded, the EXE file can be copied over to the system where Vision4D has been installed and run to perform the installation.
Using the GPU for ML Operations
Once the GPU packages have been installed, the option to use the GPU for ML acceleration will appear in the application preferences:
Note that this option may remain greyed out if the GPU drivers that are currently installed are not compatible. In that case, the user should update the drivers and restart the system. Once this option is enabled the software will default to using the GPU for any ML operation.
Compatible GPUs
This package makes use of CUDA, a proprietary technology available on NVIDIA GPUs. Please see this Wikipedia article or consult the NVIDIA documentation to find out if your GPU is supported.