[image] generation failed

[kozistr]

some kind of problems, model? or sth don't work at all...

I tried some tries for image scaling like scaling to [0,1], subtracting div2k_mean, etc.... (paper way, my way)

but all failed :( so it needs to be fixed in some way.

[kozistr]

maybe, mean_shift function referred in the paper is the problem of image generation.
currently on-fixing.

mean_shift is for image pre/post-processing.

[taniaCHY]

Can you send me a copy of your DIV2K test results?Email 1964771130 @qq.com

[kozistr]

Currently, I just failed to generate the results because of some reasons in pre/post-processing functions.

The image just collapsed(?) at global step 3000. I tried several tries that scaling image into [0, 1] or subtracting rgb_mean, etc... but all didn't work to me :( Sorry for that

[taniaCHY]

Due to time, I have no time to run the experiment of it, so If you solved this problem ，please send me the result of the test with div2k? I want to compare the results of my experiment when I write my paper

[kozistr]

Actually, It'll take some time because I'm having troubles with computing resources for running this model. (only what I got is a single GTX 1060 6GB :()

Maybe, you can use pre-trained model & benchmark from the original pytorch code which may be run perfectly.

Of course, later, I'll try to fix this issue, too!

[xliucs]

Hi,

Thanks for releasing the repo. I am also very curious what results (benchmark) did you get from this repo? I also tried another TF implementation, but did not get results as the original paper states.

[kozistr]

Hi,

Thanks for releasing the repo. I am also very curious what results (benchmark) did you get from this repo? I also tried another TF implementation, but did not get results as the original paper states.

Hi! thx for the interest!
I've been busy these days so there were no any progressions :(

but i got some time recently, so, maybe, i'll try to re-fixing and reproduce this paper!