endywon / microast Goto Github PK

MicroAST和StyTR2

F:\MicroAST-main>python test_microAST.py --content inputs/content/1.jpg --style inputs/style/shuimo2.jpg
Traceback (most recent call last):
File "test_microAST.py", line 120, in
output = network(content, style, args.alpha)
File "D:\python3.7\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "F:\MicroAST-main\net_microAST.py", line 317, in forward
style_feats = self.style_encoder(style)
File "D:\python3.7\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "F:\MicroAST-main\net_microAST.py", line 64, in forward
x1 = self.enc1(x)
File "D:\python3.7\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "D:\python3.7\lib\site-packages\torch\nn\modules\container.py", line 204, in forward
input = module(input)
File "D:\python3.7\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "F:\MicroAST-main\net_microAST.py", line 18, in forward
x = self.conv_layer(x)
File "D:\python3.7\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "D:\python3.7\lib\site-packages\torch\nn\modules\conv.py", line 463, in forward
return self._conv_forward(input, self.weight, self.bias)
File "D:\python3.7\lib\site-packages\torch\nn\modules\conv.py", line 460, in _conv_forward
self.padding, self.dilation, self.groups)
RuntimeError: Given groups=1, weight of size [16, 3, 9, 9], expected input[1, 4, 775, 511] to have 3 channels, but got 4 channels instead

Thanks for your sharing your code. It's wonderful work I think~~

I have one question about the calc_cs_loss.py. when I would like to calculate the StyleLoss or ContenLoss by using the calc_cs_loss.py, the code mentioned that cuda out of memory. My device is an NVIDIA RTX 3090 GPU (24G). May I know how can I calculate the style loss and content loss? In your paper, you just use one NVIDIA RTX 2080 GPU.

Thx for your sharing again!

what an excellent work about your StyleDiffusion. I am so interested about your work, would you please share this code?

from skimage.metrics import structural_similarity as ssim
def SSIM(stylized,content):
ssim_=ssim(stylized,content, multichannel=True)
#ssim_=ssim(stylized,content , win_size=11, multichannel=True,sigma=1.5, data_range=1, use_sample_covariance=False, gaussian_weights=True)
return ssim_

I enjoyed reading the paper and would love to play with the model. What are your plans in regards to the code release?

Hi, thank you for sharing your work.

I currently have it running video with a 1920x1080 content input (>30 fps), but I'm not really seeing a style transfer with the weights posted here. I definitely see color transfer, but it's not really picking up spatial style details.

I've also ran it just with your script for testing images, and I don't really see the style transfer there as well. Just color replacement, blurring, and some artifacts.

Am I missing something? Do I need to train before testing? If I do have to train, can you give me an idea of how long it would take with N GPUs?

Thanks,
Keith