View Single Post
Old 16th November 2015, 01:59   #34278  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by Sp00kyFox View Post
anyways, I wanted to draw some attention to a new developed downscaling algorithm by ETH Zurich which is called "Perceptually Downscaling" which gets pretty amazing results:
https://graphics.ethz.ch/~cengizo/imageDownscaling.htm

I think that could be a gread addition to madvr. according to the paper the performance of is pretty good (they implemented it with matlab) and it should be possible to execute it in realtime. it seems that they didn't publish the code though but the algorithm is explained in the paper and presented in pseudo code, it only consists of convolutions and sums.
That looks interesting, bit worried about the 'patent pending' though.

Quote:
Originally Posted by huhn View Post
not sure about this downscaler. is sharper than bicubic but terrible aliased at the same time.

just look at the hair.

the comparison algorithm are pretty weak to say it friendly.
Quote:
Originally Posted by ashlar42 View Post
I concur, if I have to limit myself at the images displayed on that page, the algorithm seems pretty terrible, to be honest. I like a sharp image, but judging from these images one is supposed to watch a Minecraft game, not a movie.

Edit: pictures here are better, though https://graphics.ethz.ch/~cengizo/Fi...ralImages.html

Aliasing is still there, but at least I can see a reason for taking it into consideration. They really ought to change the pics in the homepage.

Edit 2: these https://graphics.ethz.ch/~cengizo/Fi...udyImages.html are even better. Interesting...
Quote:
Originally Posted by madshi View Post
Could be worth a try. It does look like it tends to produce heavy aliasing in some situations, but maybe it could be improved by running some sort of smoothing filter. I don't have time for this atm, though.
The aliasing you're seeing seems to be because they've upscaled the images using point sampling. Presumably to make it easier to see differences (otherwise you'd have to look at a really small image). The other algorithms are simply too blurry to make the aliasing noticeable.

Last edited by Shiandow; 16th November 2015 at 02:26.
Shiandow is offline   Reply With Quote