How Can The Backend Runs So Fast? about sketch-to-art HOT 4 CLOSED

I suggest you took a look at their model implementations and inference code.
Seems the PyTorch version uses more VGG model during the inference stage.

from sketch-to-art.

I suggest you took a look at their model implementations and inference code.
Seems the PyTorch version uses more VGG model during the inference stage.

I don't know where is the problem, don't you use five codecs in the backend?

In the PytorchWCT:
WCT.py:

util.py:

if you would like to running the PytorchWCT in windows or Pytorch 1.3+, here MyPytorchWCT I made some changes to make the project run on windows and python1.3+, and the 8 models arethe same as yours in your releases.

from sketch-to-art.

I don't know the exact reasons why his implementation is slower.
The implementation in my repo is ported from https://github.com/eridgd/WCT-TF.
Please take more look at the source code, since all the model implementations are clear there. And you could compare the differences accordingly.

from sketch-to-art.

OK, Thx

from sketch-to-art.