Microblocs after resize

[Tueurne]

I have some microblocs after a resize 2x2 maybe or 4x4 I don't know but I am quite sensitive. Is it a bug from avisynth or not ? I try 2.52 and 2.53 no change. I haven't this problem before

This is my script :

mpeg2source("H:\Matrix\Temp\Matrix.d2v",cpu=4,idct=2).trim(30400,30800).crop(0,80,0,-80).deen("w3d",1,2,3,3,4,0.3,7).undot().bicubicresize(608,240,0,0.72).limiter()


I have no problem with mpeg2source alone but as soon as I put the resize microblocs apear.

I make a sample, microblocs appear in the wall about 12, 14s fo the most visbles

http://tn2h.free.fr/unite-video/samplemicroblocs.avi

[Caspar]

May be it's a problem with a 'over' smoothed source on some scenes and the encoder fails to allocate the right amount of bits for those frames (I am a newbie at this stuff, don't flame me if I am wrong :o ). You can try and use a filter called Blockbuster to add noise to those frames and force the encoder to allocate more bits to those frames.

[Tueurne]

This is not the same kind of blocs not 16x16 like a encode problem but some very small blocs

Did you see microblocs on your computer ?

[Caspar]

Taking a quick one pass on the sample, I can see 1 small white block at Frame 366 (0:00:14.640). So is this what you are speaking of? Otherwise, I don't see anything wrong with the sample other than it being over blurred ;) Maybe you can give me a frame number so I can take a closer look? My newbie eyes are deceiving me...

[yaz]

@tueurne
can't download the avi but ...
all filters u use can(!) produce such kinda artifacts. try a stepwise (one by one) elimination! my order would be : undot, bicubic (use lower c, say, 2/3 ... 1/3 to see its effect), edeen, mpeg2source(cpu=...) ... just my guess !!!
btw
which mpeg2 decoder do u use ?
what is edeen('w3d'...)?

the bests
y

[Tueurne]

@Caspar
This is not one bloc but some small blocs. If you don't see it maybe it come from my decoder or I don't know graphic drivers, conflics ....
blur come forum H263 matrice and devapi4. It is better with HVS good ;)

@yaz
My sript is the script I use to make the sample. But If i use only Mpeg2dec3 without resize, all is OK. As soon as I put the resize, I see problems

deen is a special dev by marc for french people very powerfull still in beta but very good this is 1.0 beta1. Maybe he will post on doom9 a day...

[Manao]

deen is a special dev by marc for french people very powerfull still in beta but very good this is 1.0 beta1. Maybe he will post on doom9 a day...For non french users, here is the link (http://atlas2.tgv.net/~media-video/forum2/viewtopic.php?t=2808) to the thread where you can download it. The readme is in french, here is an ad-hoc translation :Deen 1.0 beta 1

Deen(clip, str mode, int rad, int thrY, int thrUV,
int tthY, int tthUV, float min, float scd, string fcf)

mode: "c2d","c3d","w2d","w3d","a2d","a3d"
2d = spatial / 3d = spatial + temporal (3 frames)
rad: radius (1-4 in 3d / 1-7 in 2d)
thrY, thrUV: treshholds : luma, chroma
tthY, tthUV: temporal tresholds (3d only) : luma, chroma
min: 1 = no reduction .. 0 = maximum reduction
When using 'w' mode, it's how the weight of the pixel is reduced when
its distance is raising
When using 'a' mode, it's how the threshold is reduced with the raising
of the distance
When using 'c', it should do nothing ( no tested, it's a guess made by
me, I can't test right it now )
scd: Scene change Detector : threshold over which a 2d mode is used, in
order not to have a temporal ghosting with high motion. Higher values
means potentially more ghosting . scd=-1 allows the viewing of the value
with debugview ( only '3d' mode, it's a debug mode, I can't test it right
now so you'll have to guess what it does )
fcf: full control file (check demo.fcf.txt in the zip file). default is "",
which means no fcf. It allows an overriding of the parameters for
ranges of frame.

Finally, defaults are : deen("c3d",1,7,9,4,6,0.5,9,"")

Edited to fit the screen.

[yaz]

Originally posted by Tueurne
If i use only Mpeg2dec3 without resize, all is OK. As soon as I put the resize, I see problems ...
yep ... my guess is the too high 'c' (=0.72) in bicubic. try the range of 0.66 ... 0.33. (for the meaning of params see avs help!)
imho, 0.72 is too high for bicubic in general, in that range use 'lanczos' (if needed !).

the bests
y

[Tueurne]

Originally posted by yaz
yep ... my guess is the too high 'c' (=0.72) in bicubic. try the range of 0.66 ... 0.33. (for the meaning of params see avs help!)
imho, 0.72 is too high for bicubic in general, in that range use 'lanczos' (if needed !).

the bests
y

That's what i thought too
0.6 is better but they are still here
long time i didn't use lanczos not compressible enought but i will try maybe good n matrix ;)

thx Yaz

[Tueurne]

ok Yaz it was bicubic ;) thx
I think i will taste lanczos again ;)