View Single Post
Old 18th December 2020, 22:42   #23  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 5,345
Quote:
Originally Posted by nji View Post
If that's the case for all dedup algs (is it?) then they have to have a stop limit.
Forego quality for speed. :-(
There are various decimation filters, and decimation with automatic compensatory timecode output for VFR . Dedup falls in the latter category . I don't think there are any others like dedup for avisynth

I think understand about the "sliding" comment. There are are accruing differences each frame , n+1,n+2,n+3 that you lose comparing to the original n frame. The definition gradually changes, and you're not really comparing n+10 vs. n.

I don't think any algorithm stores a frame in memory to be able to compare a frame 10, or 20, or 100 frames down the line.


But "quality" or more accurately described "accuracy" is definitely an issue. It's difficult to measure similarity (the grey area mentioned earlier) . You can have subtle x,y translation shifts in the frame during film transfers, sometimes 1 pixel or even subpixel, that register as different, but it's clearly meant to be a duplicate. PSNR is classically fooled by that. You can have clearly duplicate frames, but grain or noise pattern added on top makes them "different" to algorithm
poisondeathray is offline   Reply With Quote