I created a Cinder block using Tensorflow’s C++ api. Could use some help filling it out a little more. I’m trying to get it up to par with Memo’s ofxMSATensorflow and would like to be able to perform imagetoimage neworks such as pix2pix or Deep Dreaming.
After weeks of not being able to build Tensorflow binaries for Windows 10 and Cuda support, I finally found this repo with prebuilt binaries…
You can download the binaries and link to the project. Make sure your env variables link to the proper Cuda. Uses Cuda 9.2.
This block uses Cinder’s OpenCV block and I customized this example to use Spout(Window’s Syphon) to stream the media in instead of loading files. This example is using a face detection model.
We may also have to link to a newer version google.protobuf for certain functions?
If you figure out a way to build your own(bazel?) binaries for Windows 10 with Cuda… let me know… If you figure out a way to make it a more robust library such as Memo’s, please share… Cheers.
1 Like
Thanks for sharing and getting a head start! Weili, one of our devs here, has also been working on getting TensorFlow for C++ up and running on Windows. He put together a post, which may have some helpful nuggets on how to get it compiled: https://medium.com/@shiweili/building-tensorflow-c-shared-library-on-windows-e79c90e23e6e
1 Like