For some advice on REVIDE...
Kayla2714 opened this issue · comments
Sorry to interrupt, I hope to apply your REVIDE dataset to my experiment, but through the experiment of MSBDN-DFF method, my PSNR is almost 1dB different from your published 22.01, so I have been troubled for a long time. I use the 384x384 crop size，100 epoches, batchsize = 4, etc. described in your paper. Finally, I lock the target in "resize the hazy inputs and GT with 3 random scales within a scale of 0.5 and 1.0". Is it completely random resize with the range of 0.5~1.0? Are The original dataset is included in the three sizes?Are 22.01 is only trained on REVIDE-indoor? And are there any other details you didn't mentioned?
I would be most grateful if I can get your help!
Hey, according to your description(1dB), your code must be wrong. Plz check
your network.
…On Sat, Apr 9, 2022 at 8:17 PM Guo09 ***@***.***> wrote:
Sorry to interrupt, I hope to apply your REVIDE dataset to my experiment,
but through the experiment of MSBDN-DFF method, my PSNR is almost 1dB
different from your published 22.01, so I have been troubled for a long
time. I use the 384x384 crop size，100 epoches, batchsize = 4, etc.
described in your paper. Finally, I lock the target in "resize the hazy
inputs and GT with 3 random scales within a scale of 0.5 and 1.0". Is it
completely random resize with the range of 0.5~1.0? Are The original
dataset is included in the three sizes?Are 22.01 is only trained on
REVIDE-indoor? And are there any other details you didn't mentioned?
I would be most grateful if I can get your help!
—
Reply to this email directly, view it on GitHub
<#1>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACQXNMO7QTQO6IU5G6FHCHLVEFYNXANCNFSM5S65YVFA>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>