To johnifehr, and everyone else.
I agree, but I still prefer MFlowFps instead of MBlockFps. The MBlockFPS gives artifacts
which makes .mkv's look squares in fast motion scenes.Most of the blocks can be
eliminated by MRecalculate, but it still doesn't give as quality picture as MFlowFPS does.
Now MFlowFPS produces different artifact: when small objects have a fast changing
background the 'ghosting' appears. I have overclocked E8500, I minimize blocksize
to 16/8 (I would prefer 8/8 though), use lsad=1400,pnew=60 (these are the penalties
for wrong vectors). So I get not as fluid, but artifact-less motions. I also lessen the
vector lenght from 4 to 3 (levels=3), since it sometimes misses good shorter vectors. By adding
chroma=false, the estimation using only Luma is being used. There is a big performance
increase, and (from my experience) no different than using both chroma&luma. I would
increase spar (pixel search) from 2 to 5, but my PC is too weak for that. Also - Reclock
is a must-have.
Sidenote: I am still wondering why nVidia, ATI or software like WinDVD, PowerDVD doesn't have motion
compensating functions for .mkv's. The newest PowerDVD has motion compensation for
DVDs and .avi's, but not for .mkv's. Even if it had this function for .mkv, I still wouldn't be able to use it
since it does not have any Video Syncing to display like MPC-HC does.
Paradox - the best things are for free? (ffdshow,avisynth,mpc-hc,reclock)
bufferback 0 ahead 8
--------------------
multinum=2
multiden=1
spar=2
pel=1
blkh=16
blkv=8
SetMTMode(2,8)
ffdShow_source()
super=MSuper(pel=pel, hpad=blkh, vpad=blkv, levels=3)
backward_vec1=MAnalyse(super, isb=true,chroma=false, blksize=blkh, blksizev=blkv, searchparam=spar, plevel=2, levels=3,lsad=1400, pnew=60)
forward_vec1=MAnalyse(super, isb=false, chroma=false, blksize=blkh, blksizev=blkv, searchparam=spar, plevel=2, levels=3,lsad=1400, pnew=60)
MFlowFps(super, backward_vec1, forward_vec1, num=FramerateNumerator(last)*multinum, den=FramerateDenominator(last)*multiden, mask=1, ml=125)
distributor()