search

jacquelinelala / GFN

Gated Fusion Network for Joint Image Deblurring and Super-Resolution(BMVC 2018 Oral)
http://xinyizhang.tech/bmvc2018/
141 stars 39 forks source link

Question about the method #23

Closed CVHW closed 2 years ago

CVHW commented 3 years ago

Dear author,

thanks for your public code. I have a question below: Did you try to use the result of the deblurring branch as the input of the super-resolution branch instead of the original input? If you employ this strategy, you could achieve a better result?

I am looking forward to your reply. Thank you.

jacquelinelala commented 3 years ago

Our ablation study shows that the concatenation of deblurring branch and SR branch(Model-1 on supplementary materials) is not as good as our dual-branch strategy due to the error accumulation. You can find the ablation study and more analysis on our paper and supplementary materials(http://xinyizhang.tech/files/supplement_bmvc2018.zip).

CVHW commented 3 years ago

Thanks for your reply.

发自我的iPhone

------------------ Original ------------------ From: xinyi zhang <notifications@github.com> Date: Mon,Feb 22,2021 10:40 AM To: jacquelinelala/GFN <GFN@noreply.github.com> Cc: Hao Wei <2294469420@qq.com>, Author <author@noreply.github.com> Subject: Re: [jacquelinelala/GFN] Question about the method (#23)

Our ablation study shows that the concatenation of deblurring branch and SR branch(Model-1 on supplementary materials) is not as good as our dual-branch strategy due to the error accumulation. You can find the ablation study and more analysis on our paper and supplementary materials(http://xinyizhang.tech/files/supplement_bmvc2018.zip).

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.