futscdav / strotss Goto Github PK

Hi,

thanks for sharing your work here! I noticed that this Implementation generates slightly blurred images when compared to other Implementations like Gatys:

This Implementation:

Gatys:

Do you know why this is the case and how this could be improved?

Thanks and best regards

i am wondering if it may be possible to get "repeatable" outputs, when locking all random stuff in the code to a seed?
or is it in strotss' nature to be random?

btw: I love this repo so much. its handy to use and easy to install... brilliant. thank you again!
Currently I am doing experiments with 1k resolution and 1000 iterations... yes takes some time, but the results are just so stunning...

Thanks for your code. I have a question that the official code optimized all steps and then progress from low sr to high sr; however, in your code, you progress from low sr to high sr in every step; is it right?
in official:
for i in range(resolution_num):
for j in step:
XXXXX
in your code:
for j in step:
for i in range(resolution_num):
XXXXX

Dear Futscdav

I faced this error while using some style images. could you please check? thanks!

Actually the images used be me are big enough. >=1024.

The following is the detail logs.

Traceback (most recent call last):

File "index-8023.py", line 169, in ai_photo
run_strotss(
File "C:\Qsync\ai\ai-tools\8023-pytorch-neural-style-transfer\strotss.py", line 451, in run_strotss
result = strotss(pil_resize_long_edge_to(content_pil, args.resize_to),
File "C:\Qsync\ai\ai-tools\8023-pytorch-neural-style-transfer\strotss.py", line 413, in strotss
result = optimize(result, content, style, scale, content_weight=content_weight, lr=lr, extractor=extractor)
File "C:\Qsync\ai\ai-tools\8023-pytorch-neural-style-transfer\strotss.py", line 355, in optimize
feat_e = extractor.forward_samples_hypercolumn(style, samps=1000) #ruanjiyang ， 寻找多少个features，缺省是1000，修改为500可以节约不少时间。效果稍微减弱。
File "C:\Qsync\ai\ai-tools\8023-pytorch-neural-style-transfer\strotss.py", line 42, in forward_samples_hypercolumn
feat = self.forward(X)
File "C:\Qsync\ai\ai-tools\8023-pytorch-neural-style-transfer\strotss.py", line 38, in forward
feat = self.forward_base(x)
File "C:\Qsync\ai\ai-tools\8023-pytorch-neural-style-transfer\strotss.py", line 29, in forward_base
x = self.vgg_layersi
File "C:\Users\qruan\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "C:\Users\qruan\Anaconda3\lib\site-packages\torch\nn\modules\pooling.py", line 153, in forward
return F.max_pool2d(input, self.kernel_size, self.stride,
File "C:\Users\qruan\Anaconda3\lib\site-packages\torch_jit_internal.py", line 267, in fn
return if_false(*args, **kwargs)
File "C:\Users\qruan\Anaconda3\lib\site-packages\torch\nn\functional.py", line 585, in _max_pool2d
return torch.max_pool2d(
RuntimeError: Given input size: (512x1x3). Calculated output size: (512x0x1). Output size is too small

Hi, is there a parameter or code change that lets the max dimension be 1024 instead of 512?

Thank you and thanks for the fun code :)

Hi，
Thanks for sharing your work！
I'm trying to understand the code，
and could you tell me what does spatial_feature_extract do?
Thanks！