-
Notifications
You must be signed in to change notification settings - Fork 623
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
SwinIR classical Image SR Training using dataparallel #84
Comments
It's not the problem of |
The system I use, do not support DistributedDataparallel. |
Thanks for your question. It's a bug because the testing image is not padded to be a multiple of window_size. Please use the latest |
Thank you sir for the update. |
See JingyunLiang/SwinIR#9 for the discussion. |
Thank you |
While going through the code changes, I failed to understand following things:
|
1, Training stage only gets 48x48 image patches as input. There is no image padding problems. 2, The patch is created in the dataloader during training. We use 48x48 patches for batch training. 3, No. We don't use it but we still keep it in the code for possible future usages. |
|
2, See Line 76 in 014142d
3, No. We just stack layers to increase the receptive field. |
Thank you sir. 3> Seems like others have done over-engineering to do hierarchical representation, which happens by default. |
While training the SwinIR classical Image SR, using dataparallel :
**python main_train_psnr.py --opt options/swinir/train_swinir_sr_classical.json**
file::main_train_psnr.py has training loop. Call to model.test(), gives error:
The header of file has comment 'training code for MSRResNet'. Seems like the file needs to be modified.
The text was updated successfully, but these errors were encountered: