search

YapengTian / TDAN-VSR-CVPR-2020

TDAN: Temporally-Deformable Alignment Network for Video Super-Resolution, CVPR 2020
MIT License
400 stars 62 forks source link

Performance question #5

Closed HarukiYqM closed 4 years ago

HarukiYqM commented 5 years ago

When I read the paper, I noticed that the PSNR reported in BD setting is better than BI setting in Vid4. It's pretty strange as BD contains degradation process. Are you sure for this? and do you have any insights about this?

YapengTian commented 5 years ago

We found that our BI model can be further improved. But, it seems that BD model is still better. A similar observation is also shown in the SIS-RDN paper. With a blurring kernel, the aliasing artifacts will be reduced during downsampling. Actually, you could try LR images from directly sampling (nearest sampling), then you might find that the SR results would be worse than results from bicubic sampling, which also has a smoothing kernel implemented in Matlab.

HarukiYqM commented 5 years ago

Thanks for your reply! Can I ask a few more questions? I wonder what is the kernel size for your BD settings and does it mean you apply BD settings to create input when you test Vid4?

YapengTian commented 5 years ago

I used the BD resizing code from the RDN https://github.com/yulunzhang/RDN/blob/master/RDN_IR/RDN_TestCode/RDN_SR_RGB/Prepare_TestData_HR_LR.m.

HarukiYqM commented 4 years ago

thanks a lot!