Thread Rating:
  • 1 Vote(s) - 5 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Super resolution
#51
Valeyard, you're right! Dumb is me Tongue
Reply
Thanks given by:
#52
I know I've brought it up before, but I don't suppose it'll hurt to leave this discussion open about using ESRGAN in upscaling.

https://github.com/xinntao/ESRGAN
Reply
Thanks given by:
#53
(2019-01-24, 02:06 PM)FrankT Wrote: I know I've brought it up before, but I don't suppose it'll hurt to leave this discussion open about using ESRGAN in upscaling.

https://github.com/xinntao/ESRGAN

As the images do not work there, for more info go there: https://arxiv.org/pdf/1809.00219.pdf

Results are interesting, but how "we" could use it? It does not upscale video, se we should "divide" it in images, and that's right... but, what it takes to upscale a single image (frame)?
Fundamental collection thread | Vimeo channel | My personal blog
Reply
Thanks given by:
#54
You'll need proper dataset for this to work properly with video and weeks of training. Pretrained stuff does not work well with low resolution video (comparable to bilinear upscale). It just needs more research in that field.
Reply
Thanks given by:
#55
Actually I want to give ESRGAN a try with some SD material. I have a feeling rendering will be very, very slow even with GPU rendering though.
Reply
Thanks given by:
#56
(2019-01-25, 02:14 AM)Valeyard Wrote: Actually I want to give ESRGAN a try with some SD material. I have a feeling rendering will be very, very slow even with GPU rendering though.

So, keep us updated! And yes, I guess it will be slooooooow...
Fundamental collection thread | Vimeo channel | My personal blog
Reply
Thanks given by:
#57
You can go ahead and just try it out yourself, it's not that hard - they've made program for this ready, but as I said, results are not great, as pretrained models aren't made for movies.

https://github.com/xinntao/BasicSR
Reply
Thanks given by:


Forum Jump:


Users browsing this thread: 1 Guest(s)