search

leftthomas / ImageDeblurring

A Keras implementation of image deblurring based on ICCV 2017 paper "Deep Generative Filter for motion deblurring"
82 stars 30 forks source link

Why the test results are still quite blurred? #23

Closed KWang1998 closed 6 years ago

KWang1998 commented 6 years ago

I ran the code with the channel rate set to 16 without changing other parts. But the output of the test images are still very blurred and not as good as the paper shows. Is there any way to improve it?

leftthomas commented 6 years ago

@WangKewei1998 you should change the channel rate as the paper said and you need a powerful GPU to run the code, because it requires many memory.

KWang1998 commented 6 years ago

@leftthomas I tried 64 as the channel rate. But it has the similar problem as other people mentioned before. The generated images of the test set are all green or blue. So I tried what they suggested and changed the channel rate to 16.

leftthomas commented 6 years ago

@WangKewei1998 maybe the problem is the loss function, I used different loss as the paper said

KWang1998 commented 6 years ago

@leftthomas So, how is your result?