Any progress with the "stress-test" video?
Pages 1
You are not logged in. Please login or register.
SmoothVideo Project → Posts by Domas4
Pages 1
Any progress with the "stress-test" video?
So that's how you programmers joke around! Good one.
Well yes, but it's too aggressive. It's either on or off. Every other scene now becomes either very fluent or very "jaggy". Wish it was more "sequential". You know, not like either on (1) or off (0), but like 0, 0.2, 0.4 , 0.6, 0.8, 1 based on how aggressive the scene is.
MAG79 excellent screenshots. Yes it gives a bit more blending area, so take a look at the bottom picture left corner, you can clearly see the black stripe being waved, which clearly spoils the picture when watching real time. The blending area might be reduced by setting plevel=0 in my script. I tried to do the same blending with SVP 3.14 by setting values same as in Mvtools2, but I could not get rid of those "black" waves. Anyhow, good that you guys are testing. Good luck
No I havn't. Here is the link for the video.
http://depositfiles.com/files/2jog7xvse
Artifacts appear on SVP with default or non-default settings, does not matter. The strange black stripes NOTE: Just turn off the frame change detection!
If you don't, it will think the frame is changing all the time and will either repeat or blend frames, and will not do the interpolation. (Something like blocks:80, scene:10000) Also turn off adaptive mode.
My mvtools script is a bit tighter. But even when I relax the settings to get smoothness I get much less artifacts than with SVP. С Пасхой!
svp_scheduler=true
svp_cache_reduce=true
svp_flow_lib="C:\Program Files (x86)\Avisynth 2.5\plugins\libflowsse.dll"
SetMemoryMax(1280)
SetMTMode(3,5)
ffdshow_source()
SetMTMode(2)
super=MSuper(pel=2, hpad=6, vpad=6, rfilter=4)
F=MFinest(super)
back=MAnalyse(super,isb=true, overlap=4, overlapv=4, lsad=1000, lambda=4000, blksize=16,blksizev=16, search_coarse=3, plevel=1, pelsearch=2, search=3, searchparam=2, badrange=-24,badSAD=8000)
forward=MAnalyse(super,isb=false, overlap=4, overlapv=4, lsad=1000, lambda=4000, blksize=16, blksizev=16,search_coarse=3, plevel=1, pelsearch=2, search=3, searchparam=2, badrange=-24,badSAD=8000)
num2=1
den2=1
Framerate<=30 && Framerate>=29 ? Eval("""
num2=2
den2=1
""") : Framerate<=24 && Framerate>=23 ? Eval("""
num2=5
den2=2
""") : Framerate<=25 && Framerate> 24 ? Eval("""
num2=12
den2=5
""") : NOP
MSmoothFps(super, back, forward, finest=F, num=FramerateNumerator(last)*num2, den=FramerateDenominator(last)*den2, algo=13, block=false, sadgamma=0.5, sadml=160, blend=true,thSCD2=155)
SetMTMode(1)
GetMTMode(false) > 0 ? distributor() : last
Or maybe I am missing something. Is there possibility to set SVP flow to mvtools2 settings? If so, I will keep trying, because I really like the adaptive method for frame interpolation.
Thanks for response.
That's what I would do Chainik, if not for the artifacts in the "Holy Motors" movie scene when the guy runs on the thread mill. I have tried every lambda, lsad, plevel, search, algorithm, refinement and etc. settings to try to eliminate it, but I could not. I think your SVPflow libraries are much more flexible than original Fizick's mvtools, but I just can't eliminate those artifacts
Dear SVP team,
Can you please adjust this file
http://www.svp-team.com/files/gpl/mvtoo … .9-svp.zip
so that I can set plevel to 0.5 ? I find 0 being too smooth and 1 being too "not smooth"
Thank you very much if you do it!
Domas4
EDIT: Got it,
Thanks for quick answer.
When using Manalyse function, the Fizick writes:
badrange: the range (radius) of wide search for bad blocks. Default is 24 (image pixel units). Use positive value for UMH search and negative for Exhaustive search.
Did he mean that I have to use negative value in searchparam or in badrange if I use exhaustive search? Like this (badrange=(-24), search=3, searchparam=1), or did he mean this (badrange=24, search=3, searchparam=(-1))
Thanks guys, admire your work with MSmoothFPS and SVP overall,
Domas4
I have noticed that 3.1.1 version does not detect scene change correctly. During fast scenes the frames are repeated (thus the stuttering), because SVP thinks that the scene has been changed. Therefore choosing "blend" in "processing of scene" changes seem to help with stuttering.
plevel=2 helps to eliminate some of the artifacts when watching 720p videos for me. I think you should reconsider using plevel=0. Just an opinion.
After using standard MvTools2, SVP 3.0 has been a breath of fresh air. The artifacts which annoyed me in the original MvTools2 are nearly gone, the new algorythms are brilliant! Finally I can enjoy 720p with nearly artifactless motion interpolation. I am not sure if the artifacts are masked or what, but I like it. The new API OpenCL also is a huge improvement in taking the load off CPU.
I am donating and so should you! Support this kind of development and it will become a big thing in the future.
Pages 1
SmoothVideo Project → Posts by Domas4
Powered by PunBB, supported by Informer Technologies, Inc.