今天在arxiv看到大神这篇文章，非常值得学习。文章中除了Fig.1展示说低高分辨率分别是16,64,实验训练中没有提起。从实验展示的图片质量来看，感觉分辨率不是很高。假设实验中的高分辨率为64的话，产生了另外一个问题：如果在高分辨率为64的x4倍进行超分对比方法实验，请问如何对DFDNet,CFPGAN,HifaceGAN这样的原始高分辨率为512的算法进行训练，或者如何跟这些算法进行比较的。

期待大神的回复。

另外，我发现4.1节中合成退化模型可能有个编辑错误，应该是LrL->LrH。

Hello! How can I get the visualized results of Fig. 3, which is about the t-SNE.

Thanks for your interesting work, would you please let me know what is the problem with this error? I have load your pretrained weight model but it does not work.
Thanks in advance

RuntimeError: Error(s) in loading state_dict for G_RLS:
Missing key(s) in state_dict: "layers_set_final.0.0.conv1.weight",.........................................

Hi, can you describe the perparation steps of test data , or is there a recommended detect method or crop size params. Thanks!

Hi all,

Is there a way to make this code work with different resolutions, e.g. 32x32 -> 128x128, or different scale factors, e.g. 16x16->128x128 (x8)? This would be helpful in comparing SCGAN with other Super-Resolution models.

Hello, thanks for your great work at first!
I 'm trying to reproduce the results in the paper. However, when I use the weights of the model trained by myself, it reports the error as follows:

Traceback (most recent call last):
File "test.py", line 59, in
main()
File "test.py", line 35, in main
a = torch.load('./pretrained_model_self/pretrained_model.pth')["G_l2h"]
KeyError: 'G_l2h'

I' d like to know what caused that. Thank you very much!

hello
i wanna run your code in GoogleColab , but there is some errors and it didn't go in the training loop when i run 'train.py' and returns this error :

usage: colab_kernel_launcher.py [-h]
colab_kernel_launcher.py: error: unrecognized arguments: -f /root/.local/share/jupyter/runtime/kernel-9e48f654-d5da-49df-8499-355aef0f7d5e.json
An exception has occurred, use %tb to see the full traceback.

SystemExit: 2
/usr/local/lib/python3.10/dist-packages/IPython/core/interactiveshell.py:3561: UserWarning: To exit: use 'exit', 'quit', or Ctrl-D.
warn("To exit: use 'exit', 'quit', or Ctrl-D.", stacklevel=1)

i downloaded the dataset and load it into the runtime storage , then upload all the python files to GooleDrive and run them step by step . it was some other errors witch resolved by searching in the internet , but i got stuck in this one .
thank you for reading this .

My env :

Name Version Build Channel

ca-certificates 2022.4.26 haa95532_0
certifi 2022.5.18.1 py37haa95532_0
cudatoolkit 10.2.89 h74a9793_1
openssl 1.1.1o h2bbff1b_0
pip 21.2.4 py37haa95532_0
python 3.7.13 h6244533_0
setuptools 61.2.0 py37haa95532_0
sqlite 3.38.3 h2bbff1b_0
torch 1.11.0+cu113 pypi_0 pypi
torchaudio 0.11.0+cu113 pypi_0 pypi
torchvision 0.12.0+cu113 pypi_0 pypi
vc 14.2 h21ff451_1
vs2015_runtime 14.27.29016 h5e58377_2
wheel 0.37.1 pyhd3eb1b0_0
wincertstore 0.2 py37haa95532_2

This is the error :

Traceback (most recent call last):
File "test.py", line 58, in
main()
File "test.py", line 35, in main
a = torch.load('./pretrained_model/pretrained_model.pth')["G_RLS"]
KeyError: 'G_RLS'

I have tried it with pytorch cuda 10.2 and cuda 11.3 just in case if that fixes the error or not,but no chance.

A solution please?

大佬你好，目前想使用相同的数据进行一些研究，因为在计算LPIPS指标时需要参考图像，不知是否方便提供。
谢谢！

My e-mail: [email protected]
Thank you very much!

Hi, Nice work! We want to you model to evalutate the LPIPS PSNR SSIM. But the high-resolution of ls3d-W and ffhq are absent.

we use your describe the method and the param in paper to generate low-resolution sample, the result is abnormal. But we use your public test data can work well. The result as follow:

the top are the result of the public test ffhq; the bottom are the result our generated samples.

can you share the high-resolution about ls3d_w and ffhq in test dataset, or the code to generate low-resolution sample. Thanks.
my email: [email protected]