Video Restoration with Enhanced Deformable Convolutional Networks

[Fraggoso]

I didn't had the time to test this yet, but it seems promising for uprezsing Videos. It's also from some of the main contributors of ESRGAN.
https://github.com/xinntao/EDVR

Someone already had time to test this with Final Fantasy Movies?

 

[satsuki]

Look great tool but can't get it works on windows (can't build some depencencies)
Will install a linux distrib in some days to test it

 

[satsuki]

Try to use it but i issue a bug (open an issue in git but posting it here too ^^)

I'm trying to use the test_Vid4_REDS4_with_GT.py script with a foder containing 98 png files (named from 00000001.png to 00000099.png).
The png resolution is 320*240.

when i use the script i got this error:

Code: [Select]

  File "test_Vid4_REDS4_with_GT.py", line 276, in <module>    main()  File "test_Vid4_REDS4_with_GT.py", line 217, in main    GT = np.copy(img_GT_l[img_idx])IndexError: list index out of range

And when i look at result i can see it have processed the first png file but not the others

Thanks for helping ^^

 

[Manakaiser]

After a quick check, Its possibly due to forced psnr calculation which requires a GT for each image.

Try creating a GT image for each frame in GT folder (just use a duplicate of the LR Frames, since its irrelevant which result the PSNR calculation gives). You might have to make the image copies in the GT folder the same resolution as your resuslts are expected to be, supposedly *4 of the LR provided you try to upscale by a factor of 4.


Let me know if it works.

If after testing it Works: You can try to avoid this step in the future by trying to remove  all lines from 215 to 273. And probably 169 to 179. But that might not work, just a suggestion if the initial advice resolves the problem for you.

 

[satsuki]

I 've done it, it's working  but i get artifact in some pictures, so no usable so far :

Seems related to model training.
I don't have time to try a model of my own but on picture without artifact seems promising as a prefilter to use before esrgan

 

[Manakaiser]

well im gald it worked but that looks indeeed relatively botched although 2 things, one its kinda interesting it doesnt artefact all frames and 2 why it would cause such drastic and more importantly locallly specific distortions; either the model provided at base is VERY specifically tailored in terms of usage sccenarios or  something else went wrong but havent used it myself so cant really provide any further insight

 

[Fraggoso]

How does it fare for pictures that don't get distortions?

 

[satsuki]

not bad, the deblurcomp model cleans well the video compression artifact without bluring the picture
good for preprocessing

 

[Fraggoso]

Hopefully you can resolve the issue.
It seams at least quite useful. and also maybe it fares better than the normal esrgan upscale picture by picture.

The squares oft logo with a lot of chocobos look horrible with esrgan. xD

 

[satsuki]

I can't solve the issue, i have artifact here and there.
I may look at this once my (with the help of manakeiser filtered files) v4 esrgan field pack will be final