I was wondering if the tensorflow plugin is limited to the use of Tensorflow as a machine learning framework? I’ve looked at the github code and to me it looks more like a python API where you can execute Python code from Unreal Engine. In that case it must be possible to use other python machine learning frameworks like theano?
Thanks
You are correct, the tensorflow plugin depends on the wonderful UnrealEnginePython plugin to do a lot of the heavy lifting, which includes a full python interpreter. There are some small modifications in the fork I use that make it a bit easier to communicate between bp<-> python, multi-threading, and including your own dependencies with a single line, but you could just default to the parent unrealenginepython plugin if you don’t need those.
Otherwise if you want to adapt the TF plugin and Theano works for your platform (e.g. windows) you should be able to modify https://github.com/getnamo/tensorflo…upymodule.json to include your dependencies and then modify your scripts https://github.com/getnamo/tensorflo…asstfpluginapi to include theano instead and largely use the rest as is (you might need to remove a few import tensorflow as tf in some scripts and will need to adapt the examples to use your theano framework)