Pretrained model link broken

[ohayonguy]

Can someone please upload the pre-trained model to a new link?
Thanks!

[Ahmadreza1365]

I cannot upload the pretrained model here. I trained it yesterday on my own device.

You can download the images from the link below and train it yourself.

https://drive.google.com/file/d/1CPMBmRj-jBDO2ax4CxkBs9iczIFrs8VA/view

Please see:

#222 (comment)

[ohayonguy]

Hi, thanks! Would it be possible for you to send it via wetransfer.com ?
Or maybe upload it to the following link?
https://drive.google.com/drive/folders/1xQ4UJzChc4WY_9_HEL6rliJ1Ck7G7MNO?usp=sharing

This would really help a lot of people here I guess. If you already trained it, then this could be great :)
@Ahmadreza1365

[Ahmadreza1365]

@ohayonguy
You're welcome.

Actually, the weights are already provided in the checkpoints folder, but I noticed that they do not work very well for the test images.

I don't have permission to upload them in the google drive you provided.
I uploaded the here.

https://drive.google.com/drive/folders/12vdMN72n0O8z9uMvPUMoTBwjUwdF9QLR?usp=sharing

you can replace it with the checkpoints folder you already have.

[ohayonguy]

@Ahmadreza1365 Okay thanks! Yes that's exactly what I encountered as well. The checkpoints provided in this github repo simply do not work well..
Thanks a lot for sharing the weights! Did you train the model according to the original implementation of the authors?

[Ahmadreza1365]

@ohayonguy I downloaded the images and divided the train and test images. 90% for train and 10% for test. Then Started the training. It took about 21 hours. My device info is:

Intel(R) Core(TM) i9-10885H CPU @ 2.40GHz 2.40 GHz
32.0 GB (31.7 GB usable)
NVIDIA Quadro T2000 with Max-Q Design

[ohayonguy]

I am still getting quite awkward results, for instance on images from the GoPro data set.

Here is the input blurry image:

And the output of the model:

Is this expected?
Thanks!

[Ahmadreza1365]

Yes, that is expected I believe. When the images are too blurry, it cannot handle them very well.

[Ahmadreza1365]

@ohayonguy
I tried this one:
https://github.com/VITA-Group/DeblurGANv2/tree/master
and the results are much better:
input:

output:

[ohayonguy]

@Ahmadreza1365 Indeed, I also tried it already.