Compare() - Example (ProCoder vs. TMPG)

[WarpEnterprises]

As an example for the now inbuilt Avisynth-function Compare here is a diagram which shows the PSNR of three encodes
(ProCoder picture type=Frame, ProCoder picture type=Field, TMPG HighestQuality)
of a short (800 fields) interlaced Hi8-clip.
Normally you can get a good hint on quality by using BitrateViewer, but Procoder uses streams which show nothing useful.

Now comes Compare (each clip against the original AVI).
It shows clearly that "field" is superior in fast moving scenes, but not optimal in low motion scenes, there "frame" is best, only in quite still frames TMPGEnc is best.

Now you can visually compare those scenes and you see it, but without Compare you won't get a hint where to begin to search.

Btw. I used the text output of Compare, put it in Excel and calculated an average of 10 frames (else you see the I,B,P-frame types very well...)

Thanks for that filter!

[attachement]

[ronnylov]

Where can I find information about how Compare works?
It's not documented in the avisynth manual at http://www.avisynth.org.
Is there a list of all built in commands (including the new ones) with description somewhere?

How do I read the diagram? Is lower or higher better?
What is the average value for all frames?
What script did you use?

[sh0dan]

Look at the documentation packages with Avisynth 2.05.

Otherwise, the 'always newest' documentation can be browsed at http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/*checkout*/avisynth2/avisynth/docs/index.html

[WarpEnterprises]

Meaning of the values:
PSNR = peak signal to noise ratio = a measure of the difference between my original AVI and the MPG (which I read in via mpeg2dec and DVD2AVI).
Therefore the higher the better (the less "noise")
An average for all frames is the sum of the points divided by the count (straightforward).

The script is like:

Ori=Avisource("Original.avi")
FRAME=mpeg2source("frame.d2v")
compare(Ori, FRAME, "", "output.txt")

You have to preview the entire clip in VD from beginning to end to get a useful textfile.

[Sharro]

Originally posted by WarpEnterprises
Meaning of the values:
...You have to preview the entire clip in VD from beginning to end to get a useful textfile.

In what mode should we do it? Full, Normal, Fast or Direct Stream?

@EDIT - It has to be full processing mode right? :-) At least it was the only one who gave me very similar results (.1 diff) to the test run with WMP6.4.2.

Is the result of any other mode (fast specialy) valid?
END EDIT

Anw I'm trying with the different modes and also with WMPClassic6.4.2

But an advice would save me some time :-))

All the best,


Sharro

[WarpEnterprises]

the mode should not matter since the info is generated by AviSynth and not by VirtualDub (which only "asks" AviSynth to work)

You must only take care that the source is COMPLETELY fetched and then closed.

[FredThompson]

Where is the attachment?

[WarpEnterprises]

many attachements are gone (there seems to be an error with the server,...)

[FredThompson]

Can you put it on your site?

[WarpEnterprises]

Which attachement do you mean? There was only a picture attached.
The filter is built-in.

[FredThompson]

yup, the picture. Your original post now only has a position marker. I've seen lots of discussion about TMPGEnc. vs. ProCoder vs. CCE but very little actual data.

I do a lot of work with DV. Have found it difficult to get ProCoder to work with DVDV2SVCD but people say it's best for DV format. That's why I'm interested in something with some data.

[Rumbah]

Hi!

In a very popular (and very good) German magazine called c't they tested mpeg2 encoders, but not with PSNR but with an analysis software called JNDmetrix-IQ (www.sarnoff.com). That's because the SNR does not represent the human way of seeing.

The results were that for VCD and SVCD TMPEG ist great.
For DVD like resolutions and bitrates (4MBit/s, 6 MBit/s, 8MBit/s), CCE beat everything with distance, in PAL and NTSC, and in DV PAL and NTSC .

For a copy of a DVD to DVD-+R, a good transcoder beats everything (Instant Copy or CloneDVD (they had a beta, 3 to 4 times faster than Instant Copy with better quality)).

Rumbah

[FredThompson]

Hmmm...then there is a big disagreement among the forum. I see people talk about the effect of codecs and always talk about old MainConcept so that is not a good comparison.

Did the magazine evaluate ProCoder? I'm only interested as people have said it is great for DV. But, if they are using outdated DV codecs...

What about CCE for CVD or SVCD? It is much faster than TMPGEnc.

Was DVD Shrink analyzed?

[Kika]

This "very good Magazine" did not write anything about the Settings of the Encoders. What did they use in TMPGEnc? CQ or CQ_VBR?
What Settings in CCE or ProCoder? They wrote nothing about that. That's not a test, that's .... (censored).
BTW.: They testet CCE SP2.65??? Oh, and they Stole a Picture, i've drawn for explaining DCT...

CCE is a lot faster than TMPGEnc, and it's a little bit better.

[WarpEnterprises]

Sorry, I don't have the pic any more.

Adding to Kika:
ProCoder is really good and really slow only for SVCD-bitrates with true interlaced source and in Mastering Quality.

My experiment on Compare is not that easy to generalize.
As I mentioned you see the type (B,P,I) of the frame, but each encoder has a different GOP structure (ProCoder even changes the GOP and other parameters DURING ONE VIDEO!) - you will compare a B-frame with an I-frame,...

So take a video (~5min), encode it with CBR, without filters and with high quality settings (search precision) and you get a good _start_ how the encoder performs.

[FredThompson]

Sigh.

OK, well,

are there are serious caveats about MainConcept DV codec 2.3.1 and TMPGEnc, ProCoder or CCE wrt:

720x480x14.95 MPEG1 at 2500 from smartdeinterlaced DV original?

CVD at 2100 from same

CVD at 2100 from DV recording/passthru of analog source, also smartdeinterlaced

Basically, I'd prefer to use CCE because of speed and DVD2SVCD support (which disqualifies MainConcept encoder) and, if any of these combinations are known to yield major problems, I'll do something different.

There are all kinds of posts that claim things like Procoder is slow but virtually noiseless at low bitrates, TMPGEnc used to lag behind in quality but is now equal, etc., etc., etc.

[Kika]

It's simple:

If Speed and good Quality is your Goal, use CCE.
If you like high Quality at low Bitrates and interlaced Video best, use ProCoder.
Do you prefer flexibility, use TMPGEnc.

[FredThompson]

That's the gist of what I've read. I assume 2100 or so qualifies as low bitrate. From what I've read, ProCoder is less prone to mosquito noise. How about CCE vs. ProCoder for progressive?

Although, having typed that, the odds are CVDs will be viewed with a proper DVD software viewer (WinDVD, PowerDVD, etc.) or on a TV so deinterlacing might NOT be a good idea. It would add to the processing time and change the nature of the original quite a bit. MPEG2 supports interlacing...

OK, having answered my own question, it appears, how about this one:

How do TMPGEnc, CCE, and ProCoder compare on speed? I've watched DVD2SVCD make a 47 minute CVD with TMPGEnc CBR with defaults in about 2 hours. 2-pass CCE was even faster. How does ProCoder compare?

[Rumbah]

Everything was encoded interlaced.
In DV Procoder was as good as CCE. They used the standard matrices and GOPs, set quality to max and for CCE, they disabled the filters. They did not measure the time, only quality.

They also tested DVD Shrink, and for only transcoding the film, it is very good, but transcoding the whole DVD (with extras), it was only ok (in American school system a C I think).

Rumbah

PS: You can say what you want, but the c't is the best German computer magazine.

[Kika]

You can say what you want, but the c't is the best German computer magazine.

Yes, but that say's a lot about German computer magazines... :devil:

set quality to max and for CCE

Yes, in a version, that's not official. The actual version is 2.67 beta or 2.66 - with different settings!
And: CCE has 2 ways to set Quality.


@FredThompson
Encoding Speed:
CCE ist the fastest. From 1.5 up to 6 times faster than TMPGEnc.
Speed on TMPGEnc and Procoder ... hm, in most settings, TMPGEnc is faster, but that depants on the settings and on the Film (interlaced or not? Noisy?)