abdullah-abuolaim / defocus-deblurring-dual-pixel Goto Github PK

Hi Abdullah!

Thanks for your wonderful work!

in ./DPDNet/config.py, it's mentioned as follows：
# mean value pre-claculated
src_mean=0
trg_mean=0.

Is it necessary to calulated the two values when I retrain the model?

Thank you and look forward to your reply!

Best,
Cheng

hello Abdullah:
Thanks to your great work!
I have some question when i test use your code and pre-trained weight, i was wondering how many gpu did you use at test stage?
(i use three RTX3090, but the result is not satisfactory, psnr is only 9.1!!!)

looking forward to your reply!

Excuse me, when i want to use 256 patch for trainning, is the overlap is also 60%?

Hello,

I'm trying to use the google pixel data that was shared in this repo and I wonder why they are gray and so dark..

Thank you very much!

The paper report the metrics of indoor and outdoor scenes， could you provide the lists of indoor and outdoor scenes in the testing set?

Thanks for this great dataset and hard work
I am also trying to create a similar but small dataset like you but for a specific domain. But I have some couple questions I still didnt get it : ) May be I am missing something.

Did you take the photo of the same scene with 2 different settings? (aperture 22 for target, aperture 5.6 for blur images(L and R) )
if this is the case,
how did you take the photo of the same scene (especially in outdoors) with 2 different settings. Did Canon have some sort of settings for that which takes photos at the same time with different settings?
if this is not the case,
did you take photos with aperture 22 or did you take photo with aperture 5.6 to extract target and sources

I would be very very appreciated if you direct me to the right way.
Thanks in advance

hi， thanks for your great work。

I want to test your model with my own images, however, i get black output for prediction, do you know what happen?

thanks

Hi Abdullah!

Thanks for your wonderful work!

In this paper, you tested the DMENet on your dataset for comparation.
Could you share the code about it.

Thank you and look forward to your reply!

Best,
chanwental

Hi Abdullah!

Thank you for the wonderful work! The experiments are really thorough and impressive!

Could you tell me when do you plan to publish your datasets and code?

Thank you and look forward to your reply!

Best,
Qi

i cant understand why activate function is sigmoid in last layer？

I am curious about why loss function is MSE？

When we want to test the 1920 * 1080 image, the program shows that the input does not match

Hi Abdullah,

Thanks again for your fabulous work, and
congrat on your acceptance on ECCV2020!

Is it possible for you to share checkpoints for models trained on 8 bit images for both DPDNet and DPDNet_single?

Best,

First, thanks for your kind work.

When experimenting with the pretrained model you provide, the average psnr result of all test images is the same as reported in the paper.
However, I don't know why the psnr results for the indoor/outdoor set are different from those reported in the paper.

Dear Dr. Abdullah:

Hi. I have read your paper and thanks for your work.
I'd like to ask how to test your model on a single image (as I noticed in your ECCV 2020 paper). I'm also doing some research on defocus image deblurring and I hope to use your method as a reference in my paper. Thanks.

BR.

Thanks for sharing the dataset, but I can not unzip the download zip file.
The dd_dp_dataset_canon.zip file I download is about 15.8GB. I use unzip dd_dp_dataset_canon.zip, I got Archive: dd_dp_dataset_canon.zip End-of-central-directory signature not found. Either this file is not a zipfile, or it constitutes one disk of a multi-part archive. In the latter case the central directory and zipfile comment will be found on the last disk(s) of this archive. note: dd_dp_dataset_canon.zip may be a plain executable, not an archive unzip: cannot find zipfile directory in one of dd_dp_dataset_canon.zip or dd_dp_dataset_canon.zip.zip, and cannot find dd_dp_dataset_canon.zip.ZIP, period.
Would you provide some advices?

Hi, good research. Looking forward to the code and pretrained model!

Thank you for your interesting and excellent work!