Importing tensorflow on Windows10 (2019)

Hi,
I succeeded to import Tensorflow and implemented arbitrary style transfer and PoseNet and so on last year.
(Sorry for submitting this too late)

my implementation environment is:

  • OS: Windows10
  • TouchDeisigner: 2019 (19930)
  • python: 3.5.6
  • tensorflow-gpu: 1.13.1
  • numpy: 1.16.4
  • scipy: 1.1.0
  • CUDA 10.0
  • cuDNN: 7.4

See arbitrary style transfer implementation video on youtube:

See PoseNet implementation video on youtube:

1 Like

Great stuff. What GPU were you using? Just wondering what power was needed for those frame rates.

Hi
Good job !
Did you use it directly in TD or thanks spout and running in python separately ?
For the style transfer which github project ?

Thanks

Thanks.
In arbitrary style-transfer,
GeForce 1080Ti(MaxQ): 10 - 12fps
GeForce 1080: 14 - 17fps
But analysis & output resolution is only 360p :sob:

In PoseNet,
GeForce 1080Ti(MaxQ):60 fps, but analysis resolution is only 360p as well.

I imported TensorFlow in TD directly, but when opening the project we have to wait about 30 seconds to import.
I used OP Execute DAT to analyze style & content, and then output using Spout-for-Python.

I do hope the operator to output OP Execute DAT processed pixel directly :blush:

Github project: https://github.com/magenta/magenta/tree/master/magenta/models/arbitrary_image_stylization

1 Like