Welcome to the TensorFlow extension for QuPath!
This adds support for loading some pre-trained TensorFlow models into QuPath using TensorFlow for Java.
It is intended for the (at the time of writing) not-yet-released QuPath v0.3, and remains in a not-quite-complete state.
Its previous main use was to run StarDist nucleus identification, although the new QuPath StarDist extension does not require that TensorFlow is available.
Important! TensorFlow Java does not currently support Mac computers with Apple silicon.
There is no pre-built version of the TensorFlow extension at this time, because most people shouldn't need it and there are lots of different permutations of dependencies that might be required for different platforms.
For that reason, it needs to be built from source. If you want to match to a specific version, you can download the source from the releases page.
You can build the extension with
gradlew clean build copyDependencies
The output will be under build/libs
.
clean
removes anything oldbuild
builds the QuPath extension as a .jar file and adds it tolibs
copyDependencies
copies the TensorFlow dependencies to thelibs
folder
Alternatively, you can create a single .jar file that contains both the extension and all its dependencies with
gradlew clean shadowjar
The default build process will use TensorFlow for the CPU.
As described at http://github.com/tensorflow/java there are alternative dependencies that include GPU support and/or use mkl.
To use any of these, add the platform to any of the building tasks above. For example, to create a single GPU-friendly jar, use
gradlew clean build copyDependencies -P platform=gpu
The platforms available at the time of writing are mkl
, gpu
, mkl-gpu
.
Not all options are available for all operating systems. For example, GPU support is not available with macOS.
When using platform=gpu
, you will need
- an NVIDIA GPU
- CUDA and cuDNN
Installation may be simplified if you include
gradlew clean build copyDependencies -P platform=gpu -Pcuda-redist
to download the required CUDA files via JavaCPP.
Before using this option, please check https://github.com/bytedeco/javacpp-presets/tree/master/cuda for the terms of license agreements for NVIDIA software included in the archives.
Warning! At the time of writing, the CUDA version used with TensorFlow Java differs from that
used with OpenCV via JavaCPP. This is likely to cause problems if trying to use both.
The extension + its dependencies will all need to be available to QuPath inside QuPath's extensions folder.
The easiest way to install the jars is to simply drag them on top of QuPath when it's running. You will then be prompted to ask whether you want to copy them to the appropriate folder.