@scharfis_brain just did a test on
one of my DV clips. In short my results from above
almost correlate with "real" footage. I will admit that this footage isn't a car crash, in that there isn't a lot going on.
Results
33.35016: (blksize=4, overlap=2, divide=0, idx=1, dct=4, truemotion=true, pel=2, search=0, searchparam=0, lambda=2700, sadx264=7, pnew=9)
33.48795: (blksize=8, overlap=4, divide=2, idx=1, dct=4, truemotion=true, pel=4)
33.49406: (idx=1, pel=4)
33.60581: (blksize=16, overlap=8, divide=2, idx=1, dct=4, truemotion=true, pel=4)
33.98558: (blksize=8, overlap=4, divide=2, idx=1, dct=4, truemotion=true, pel=2)
34.02027: (blksize=8, overlap=4, divide=2, idx=1, dct=1, truemotion=true, pel=2, search=3, sadx264=7)
34.02353: (idx=1, dct=1)
34.03998: (idx=1)
34.04628: (idx=1, dct=4)
34.05154: (blksize=16, overlap=8, divide=2, idx=1, dct=4, truemotion=true, pel=2, search=3, searchparam=0, lambda=2700, sadx264=7, pnew=9)
34.05336: (blksize=16, overlap=8, divide=2, idx=1, dct=4, truemotion=true, pel=2)
34.07283: (blksize=16, overlap=8, divide=2, idx=1, dct=0, truemotion=true, pel=2, search=0, searchparam=0, lambda=2700, sadx264=7, pnew=9)
34.07558: (idx=1, dct=1, search=3)
34.08640: (blksize=16, overlap=8, divide=2, idx=1, dct=4, truemotion=true, pel=2, search=0, searchparam=0, lambda=2700, sadx264=7, pnew=9)
34.09980: (blksize=16, overlap=8, divide=2, idx=1, dct=1, truemotion=true, pel=2, search=3, sadx264=7) - Very SLOW
Code:
AVISource("DV-Cam-Test.avi").ConvertToYUY2(interlaced=true)
SeparateFields()
SelectEven()
BicubicResize(320,240)
b=last
SelectEven()
vectors=mvtools_MVAnalyseMulti(blksize=16, overlap=8, divide=2, idx=1, dct=4, truemotion=true, pel=4)
bw1_vec1=vectors.mvtools_MVMultiExtract(0)
fw1_vec1=vectors.mvtools_MVMultiExtract(1)
#~ bw1_vec1=mvtoolsOLD_MVAnalyse(isb=true, blksize=16, overlap=8, divide=2, idx=2, dct=4, truemotion=true, pel=2)
#~ fw1_vec1=mvtoolsOLD_MVAnalyse(isb=false, blksize=16, overlap=8, divide=2, idx=2, dct=4, truemotion=true, pel=2)
c=mvtoolsOLD_MVFlowFps(last, bw1_vec1, fw1_vec1, num=30000,den=1001)
Compare(c, b, logfile="u.txt")
StackHorizontal(b)
StackVertical(StackHorizontal(Overlay(b,c,opacity=0.5), StackComparer(c,b,"")))
Trim(2,framecount()-2)
SelectOdd()
Function StackComparer(clip x, clip y, string s)
{
sb1=Subtract(x,y)
sb1=Overlay(sb1.Levels(128,1,130,0,255),
\ sb1.Levels(126,1,128,0,255).PointResize(320*4,240*4).ConvertToYUY2().Invert("y").ConvertToRGB32().PointResize(320,240)
\ ,mode="add").Levels(34,1,255,0,255).Subtitle("Black pixels means less/no change")
sb1
}
Load the txt file into excel and do an average on the PSNR column. The odd thing is pel=4 seems to be worse then pel=2 and large blocks are better.