CQ vs. CQ_VBR ... Very Interesting... - Page 23

Boulder · #**441** 01-10-2003, 01:59 PM

Quote:

Originally Posted by kwag

Quote:

Originally Posted by Boulder

Kwag,

which motion search precision have you been using in your latest tests?

If it's 'high quality', would you please test both 'high quality' and 'motion estimate search' and compare them against each other. I'd really like to know if it's worth switching to the much slower one.

Hi boulder,

I've been using "High quality". I'll make some tests today with this new stuff and "Fast" motion search

Edit: Test finished on a sample. "High quality" is FAR better than "Fast" motion estimation. At the same CQ value, the "Fast", produced a sample slightly larger than the "High quality" sample. And the quality is also slightly inferior. So if I lower the CQ value to match the size of the "High quality", the result will be even worse. So, no contest, "High quality" wins

-kwag

Thanks Kwag, looks like I'll be switching to HQ then

acidfire · #**442** 01-10-2003, 02:23 PM

@kwag
I'm sure everyone is waiting for your new kvcd template and have already fallen off their chairs waiting

. And the way it sounds your getting there.

Keep up the great work.

black prince · #**443** 01-10-2003, 02:55 PM

@Boulder,

With Tmpgenc --> Settings --> Video Settings try setting
P_Picture Spoilage = 0 and B_Picture Spoilage = 0.
This I understand (as SansGrip says) improves file size. I
don't know why, but when I tried this and it worked.

-black prince

Boulder · #**444** 01-10-2003, 03:18 PM

Quote:

Originally Posted by black prince

@Boulder,

With Tmpgenc --> Settings --> Video Settings try setting
P_Picture Spoilage = 0 and B_Picture Spoilage = 0.
This I understand (as SansGrip says) improves file size. I
don't know why, but when I tried this and it worked.

-black prince

Thanks for the tip. I'm currently trying to figure out whether I should use CQ + the latest GOP + beta-1a matrix or simply CQ_VBR + 2-B-frame-GOP + beta-1 notch matrix for 1CD at 352x576. I encoded "Bend It Like Beckham" with CQ but didn't really like the results, there were ugly blocks nearly everywhere. It's a 2-CD DivX rip originally, but I've had very good results with similar sources earlier. Onto the testing it is then..

SansGrip · #**445** 01-10-2003, 05:21 PM

Quote:

Originally Posted by Boulder

didn't really like the results, there were ugly blocks nearly everywhere

Welcome to CQ mode

. It's perfect for some movies -- for whatever reason the blocks just don't seem to show up. But on others, it's really not good at all. At the moment I do two sets of samples, one with CQ and one with CQ_VBR+Blockbuster, then use whichever is best.

kwag · #**446** 01-11-2003, 01:48 AM

Hello everybody,

Here's another short sample (extremely high contrast scene) from K-Pax, made with the new parameters: http://www.kvcd.net/kpax-newgop-newmat.mpg (no audio)

K-Pax is 2 hours and 40 seconds. That's what it looks like on one CD-R at 528x480 with audio at 112Kbps. CQ used was 71.3.

-kwag

SansGrip · #**447** 01-11-2003, 09:41 AM

Quote:

Originally Posted by kwag

Here's another short sample (extremely high contrast scene) from K-Pax, made with the new parameters

Looks really good, though I do notice a little blockiness on Kevin Spacey(tm)'s face, and a slight "pulsing" in the background and on his forehead.

I'm not sure we'll ever be able to get rid of that pulsing completely, since if you single-step through each frame of any MPEG-1 stream you'll see that the I-frames are much blockier than the P- and B- frames, which is why we get a pulse every time the I-frame comes around. You can see it even on completely standard VCDs. That's one area where MPEG-2 wins -- its I-frames are much less blocky.

I'd be really interested to see the same clip done with CQ_VBR and Blockbuster method="noise" (maybe variance=0.7 or so). Maybe that would get rid of the little bit of blockiness, but how much Gibbs would be introduced?

kwag · #**448** 01-11-2003, 12:20 PM

Quote:

Originally Posted by SansGrip

I'd be really interested to see the same clip done with CQ_VBR and Blockbuster method="noise" (maybe variance=0.7 or so). Maybe that would get rid of the little bit of blockiness, but how much Gibbs would be introduced?

Here's the CQ_VBR version: http://www.kvcd.net/kpax-newgop-newmat-cqvbr.mpg

Both clips were made with this script:

Code:

LoadPlugin("C:\encoding\MPEG2DEC.dll")
LoadPlugin("C:\encoding\fluxsmooth.dll")
LoadPlugin("C:\encoding\blockbuster.dll")
LoadPlugin("C:\encoding\legalclip.dll")
LoadPlugin("C:\encoding\sampler.dll")

Mpeg2Source("K:\KPAX\VIDEO_TS\movie.d2v")
LegalClip()
BilinearResize(496,256,8,57,704,360)
FluxSmooth()
Blockbuster(method="noise", variance=.3, seed=1)
#AddBorders(16,112,16,112)
LegalClip()

#Sampler(length=24)
## MPEG size = ((Total frames/MovieTimeInMinutes)/24) * MPEG sample file size ##

To me, it looks like

-kwag

SansGrip · #**449** 01-11-2003, 02:22 PM

Quote:

Originally Posted by kwag

Here's the CQ_VBR version

A little more Gibbs, but slightly fewer blocks.

Quote:

To me, it looks like

What does

mean?

kwag · #**450** 01-11-2003, 02:57 PM

Quote:

Originally Posted by SansGrip

What does

mean?

SansGrip · #**451** 01-11-2003, 03:35 PM

Quote:

Originally Posted by kwag

Oh hehehe. You really think it looks that much worse?

kwag · #**452** 01-11-2003, 03:43 PM

Quote:

Originally Posted by SansGrip

Quote:

Originally Posted by kwag

Oh hehehe. You really think it looks that much worse?

No, just a little worse

I don't know how it will look at 352x240. I haven't tested that yet. But I saw K-19 on my HDTV and it looks by far better than every previous matrix or GOP

So for 528x480, New GOP + CQ + BETA-1a Notch Matrix wins by far

-kwag

SansGrip · #**453** 01-11-2003, 04:06 PM

Quote:

Originally Posted by kwag

So for 528x480, New GOP + CQ + BETA-1a Notch Matrix wins by far

Ok... I shall give it a quick test then make that my default TMPGEnc template

.

black prince · #**454** 01-11-2003, 04:20 PM

@Kwag,

Kwag wrote:

Quote:

I don't know how it will look at 352x240. I haven't tested that yet

Just finished testing LBR using GOP 1-12-1-1-24, Q-Matrix Beta-1a,
CQ_VBR=60. Here's my avs script:

mpeg2source("D:\Temp\movie.d2v")
LegalClip()
#
GripFit( source_anamorphic = true, dest_anamorphic=false, width=352, height=240, overscan=1)
#
opening=Trim(0,3093)
opening=opening.TemporalSoften(3,8,30)
#
movie=Trim(3094,145515)
movie=movie.Blockbuster(method="noise", variance=.5, seed=1).FluxSmooth()
#
credits=Trim(145516,0)
credits=credits.Greyscale().TemporalSoften(3,8,30)
#
last=opening+movie+credits
#
LegalClip()
#
#Sampler(length=24)

At 352x240 it's very hard to detect flashing unless I got right up
to my PC. At higher resolutions (e.g. 528x480) it more noticable.
There was hardly any noticable blockiness probably due to Blockbuster
noise. The picture was softer looking so that Gibbs noise wasn't noticable.
I wanted to sharpen the picture just a little without dramatically increas-
ing file size, anybody have suggestions. From 5+ feet the picture is
very good. I'd say there was an improvement from the older GOP and
standard Q-Matrix. File size with the new GOP and Q-Matrix has
decreased

-black prince

kwag · #**455** 01-11-2003, 04:24 PM

Quote:

Originally Posted by black prince

At 352x240 it's very hard to detect flashing unless I got right up to my PC.

I notice that viewed on a standalone player, this flashing is almost negligible.

-kwag

KingTuk · #**456** 01-11-2003, 04:27 PM

So is the template ready to release to the masses?

or

Does more testing need to be done?

jorel · #**457** 01-11-2003, 04:30 PM

Quote:

Originally Posted by KingTuk

So is the template ready to release to the masses?

or

Does more testing need to be done?

...to the masses?

yes,i want to know too!

black prince · #**458** 01-11-2003, 04:37 PM

@Kwag,

Kwag wrote:

Quote:

I notice that viewed on a standalone player, this flashing is almost negligible.

It looked the worst when I tested 704x480 with CQ. I could stand back
a few feet and still see flashing. SansGrip has a point about MPEG-2,
that there are really two compressions taking place for I, P, and B
frames. The blocks are compressed as well as the entire frame itself.
Your latest adjustments to GOP and Q-Matrix has made a noticable
decrease, which leads me to believe there still possibly a solution.
At high CQ or CQ_VBR, Gibbs noise seems to be almost gone.
If it weren't for flashing this would be equal to DVD quality.

-black prince