justanhduc / brushstroke-parameterized-style-transfer Goto Github PK

Great job reimplementing in pytorch! However, In the case of the same number of iterations(1000), I found the result different from the official implementation(And I see the parameters used in your code are the same as the official version). Are there any changes in your code that affect the final result? Images are as follows.

Looking forward to your reply.

offical 1000 iterations

pytorch 1000 iterations

Q1, In Algorithm 1, D_{strokes} has shape hwn and w_{strokes} has shape n, then there can't be max across axis 4. Or I guess it should be D (shape hwn*s) rather than D_{strokes} if there is a max across axis 4.
Q2, What are the temperature values did you use for sigmoid and softmax?
Q3, where is the corresponding code of Q1?

Hello justanhduc!
When I run the code 'python nst.py' on linux after installing the Neuralnet-pytorch, there's an error 'PermissionError: [Errno 13] Permission denied: '/ssd2''. How can I deal with the bug?

Hello, I installed torch_cluster according to your requirements, but in torch_cluster, but the following problems occurred, can you help me solve it?
Traceback (most recent call last):
File "F:/brushstroke-parameterized-style-transfer-master/brushstroke-parameterized-style-transfer-master/main.py", line 11, in
from param_stroke import BrushStrokeRenderer
File "F:\brushstroke-parameterized-style-transfer-master\brushstroke-parameterized-style-transfer-master\param_stroke.py", line 3, in
from torch_cluster import knn
File "F:\brushstroke-parameterized-style-transfer-master\brushstroke-parameterized-style-transfer-master\torch_cluster_init_.py", line 14, in
torch.ops.load_library(importlib.machinery.PathFinder().find_spec(
AttributeError: 'NoneType' object has no attribute 'origin'

Process finished with exit code 1

great job reimplementing in pytorch, does it improve cpu inference runtime? the original took 3 hours on CPU (Intel(R) Xeon(R) CPU E5-2630 CompVis/brushstroke-parameterized-style-transfer#3