MPEG2 Encoder test

[manono]

Hi-

...and put at the end of your script:
'ConvertToYUY2()'

This is interlaced DV footage, right? Then it had better be:

ConvertToYUY2(Interlaced=True)

shouldn't it?

But I agree that just dropping an RGB AVI into CCE could be the source of the color problems, since the CCE colorspace conversion routines aren't nearly as good as those of AviSynth.

grill1968, if you already have AviSynth installed, the script could be a simple as:

AVISource("C:\Path\To\Video.avi")
ConvertToYUY2(Interlaced=True)

Create a new .txt file, but with the .avs extension instead, named something like, "Video.avs". Inside you'd have those 2 lines (modified for the location of the video). After testing in VDubMod to make sure it opens OK, you'd drag it into the CCE window.

[hrlslcbr]

ConvertToYUY2(Interlaced=True)

Yep, me wrong
As I said; I've compared source with encoded file many times, and I don't get any colour difference like that.
It must be CCE that's doing something to the RGB file.

[grill1968]

Thank you guys, the colors have got closer to the original. I've already learnt of this script among very few which I've had a knowledge of, but I didn't know if it would be useful in my situation. Now I will concentrate on the quantizer matrices as manono suggested.

[manono]

Hi-

I was doing some more thinking (always a dangerous thing), and was wondering if perhaps Procoder uses a lower bitrate or softer matrix than does either CCE or TMPGEnc. If so, that could account for its less amount of noise. You could check its matrix by opening a Procoder encoded MPV, M2V, or Vob in ReStream. So, in addition to perhaps trying out a high bitrate matrix as described earlier, you might also try out a lower bitrate matrix. CCE has quite a good one included. It's the Very Low Bitrate Matrix, so no typing in and saving of a new matrix would be required. If it turns out that Procoder uses a matrix all its own, you could also add that to the CCE list of matrices, and then see if CCE using the Procoder matrix produces the same pleasing results as does Procoder. I'm full of bright ideas for you to try out.

I've never used Procoder myself, but unless it's doing some filtering on its own (a check of the settings should tell you-you'd be looking for such things as a Vertical Filter or a Low Pass Filter), I would think it would have to boil down to the matrix used.

[hrlslcbr]

...the colors have got closer to the original...

Good to hear that. Can you post a pic?

[The_Doman]

I've never used Procoder myself, but unless it's doing some filtering on its own....

I am using only procoder and I am pretty sure it does some internal filtering.
But the big plus is it produces always great results without the need to fiddle around so much with the settings...

[Wile_E]

Procoder supposedly has a low-pass/2D filter that it uses. It is always on, and can't be adjusted. CCE also has one, but it is adjustable. I have CCE Basic, and you turn it on by checking "Quality Setting". Then adjust the "simple/complex" slider. Moving it toward "complex", adds more 2D filtering. According to an email I got from CinemaCraft support, it is an adaptive 2D spatial filter. I usually have it set one tick from the right, so the "quantization characteristics" shows 12.

[manono]

That's good to know, Wile_E. Thanks to both you and The_Doman for the posts. Me, I turn off the Low-Pass Filter in CCE, because I don't want my DVD backups filtered, especially not by a spatial filter. I can understand how people dealing with DV footage might want it on, though.

You also seem to be confirming that it's possible to emulate Procoder encodes in CCE by strengthening the Low-Pass Filter settings. I would also guess that their matrix plays a part, although I could be wrong. I wouldn't mind getting a look at the Matrix that Procoder uses. Just open a Procoder encoded MPV, M2V, or Vob in the latest version of DGIndex, go Options->Log Quant Matrix, then File->Save Project, and copy and paste from the quants.txt file it creates.

[grill1968]

manono:

I think the point is the picture/filter settings rather than the quantizer matrices. As you suggested I got the Procoder's non-intra matrix with ReStream. Yes, Procoder uses its own non-inta matrix (intra: standard) but it's almost exactly the same as compared to that of CCE standard:

Using Procoder's matrix in CCE 2.70 no quality improvement was achieved for my test footage. CCE's very low bitrate matrix gave a terrible result. I'm pretty sure there should be an optimal combination of filters/setting in CCE or TmpgEnc which beats Procoder in my test. But this trial-and-error game needs much knowledge and time.

While_E, manono thanks for your hints, I'm trying to get more in CCE guide. Yes, I also guess so we should focus on low-pass, anti-noise, blur etc. filter settings.

[manono]

Yep, you're right. Although that non-intra matrix does filter more than the CCE Standard, it's not what I expected. So I would guess that the difference is in the Low-Pass filtering. If you're not tired of the game already, you might next experiment with the Low-Pass filter settings that Wile_E suggested. But whether you do or not, thanks very much for taking the time and effort to run the tests.

[vhelp]

Yup, I too have being on the same belief for a while now as well.

It is definately filtering. But, I would have to say, a damn good
one at that. When the source is clean mostly noiseless.. ie, dv cam
footage; dvd rips; or clean cable/satellite; it produces similar
MPEG qualities as my SuperBIT dvd movies such as one I have,
in front of me.. "The Fifth Element". The quality are almost non-
mistable.

As for the last matrice posted above, that one is not a special one.
It is the defualt one used in TMPGenc and CCE. In fact, I performed
a few encoding tests (to be sure) on Interlace and Telecine sources,
and the Matrices were the same through-out.. using the defaults that
TMPGend and CCE uses.

There is *definately* something unique inside Procoder's performance.
It is my opinion that as long as your source is *clean* it will return
near perfect results.. w/ small exception of the colorspace difference.
It just will not re-produce the original source's. The only MPEG
Encoder that will do this, is TMPGenc. But, asside from this color
issues, though very small, this is probably the best Encoder out there.

What will make Procoder (PC) better though, is when your source is
clean, and also, if color smoothness is under control. I've had a
lot of time with PC, and I found that if color smoothness is even,
throughout the video frame, it will do an outstanding job on the final
MPEG.. yielding similar results to that SuperBIT dvd I mentioned above.

The secret would seem to be.. clean source and color smoothness or
evenness.

So, what is this color smoothness/evenness that I'm on about ??

This is when you capture a given source, and the color's smoothness
or balance or whatever you technically refere it as, is showing a
slight blend of Blue and Green, caused by chroma blending or YUV
procesing. I can't quite put my finger on the spot, but based on
what I have seen and done, this is about the best I can do in describing
the phenomina that I am on about here.. then again, I could be wrong.

Matrices ...

Obviously, they have their place. But more so in concerto w/ the color
smoothness and video cleanlyness. But, there is also one last key
element to this puzzle. The source's color level or brightness level..
seems to have an adverse effect on the Matrice's peformance. Further
testing will prove where this will ultimately lead up to

In my analysis with the above, Procoder v1.5 was the Encoding Tool.

-vhelp 3562

[grill1968]

vhelp: Thanks for your valuable input.

...As for the last matrice posted above, that one is not a special one. It is the defualt one used in TMPGenc and CCE. In fact, I performed a few encoding tests (to be sure) on Interlace and Telecine sources, and the Matrices were the same through-out.. using the defaults that TMPGend and CCE uses.

That's not the exactly same to the standard matrix used by CCE 2.70, please take a close look at the last 4 numbers in the row 5. But it doesn't really matter.

[manono]

And I thought that I sometimes wrote long posts, vhelp, but you're the king.

Although its generally agreed that CCE has better quality output for progressive sources (although maybe not by you), I had always wondered why Procoder was considered to have better output (and sometimes much better output) for interlaced sources. I had originally thought it was the matrix used. But that's been proven wrong now. So, if all three of the major encoders have similarly good engines, and Procoder has better quality for interlaced sources, then something else is at work, and I also now guess it has to be the filtering. If it's given that you need a much higher bitrate for the same quality encoding noisy interlaced sources, when compared to progressive encoding of clean sources, then encoding noisy sources, or encoding interlaced sources at relatively low bitrates (grill1968 was using 4000 for his tests), to keep such good quality, Procoder uses a Low-Pass Filter, or a Vertical Filter, or both, to improve on the output. But this filtering comes with a price. That price is loss of detail and sharpness. If the source is noisy or interlaced, or of low quality to begin with, or you're using a relatively low bitrate, that may be a small price to pay. But if your source is clean and progressive, then that's way too high a price, for me anyway.

Matrices ...

Obviously, they have their place.

For me recently, they not only have their place, but are in first place when I set up CCE for encoding. I've never looked at the matrix that Superbit DVDs use. Maybe it's this one:

Intra Matrix:

08 08 09 11 13 13 14 17
08 08 11 12 13 14 17 18
09 11 13 13 14 17 17 19
11 11 13 13 14 17 18 20
11 13 13 14 16 17 20 24
13 13 14 16 17 20 24 29
13 13 14 17 19 23 28 34
13 14 17 19 23 28 34 41

NonIntra Matrix:

08 08 08 09 09 09 09 10
08 08 09 09 09 09 10 10
08 09 09 09 09 10 10 10
09 09 09 09 10 10 10 10
09 09 09 10 10 10 10 11
09 09 10 10 10 10 11 11
09 10 10 10 10 11 11 11
10 10 10 10 11 11 11 11

This is the matrix in general use on the best DVDs put out by the best DVD production companies. You'll notice it's very different from the Standard Matrix, with much less high frequency detail being filtered away. If you're using Procoder, with its Standard Matrix and additional filtering, or even CCE at default settings, which also does additional filtering, then the loss of detail should be quite noticeable when compared to the original DVD, or compared to using this matrix and turning off CCE's default filtering. Obviously it's not always, or even often, possible to use the same matrix as on the original, when compressing down to DVD5. But you can use a similar one which compresses a bit better. I might suggest studying the matrices used on your favorite DVDs, and then experimenting with their use in your tests. The improvement might not be so noticeable in close-ups, but for mid and long-range shots, you should notice an improvement in the detail and sharpness retained when compared to the use of the Standard Matrix. Turning off all filtering in CCE should make for even greater, though less noticeable, improvement.

[dipstick]

No doubt about it, Procoder uses filters to reduce noise, gain color sturation and add contrast. At least that's my assumption.

Procoder would be my first choice for noisey video such as shot in low light. For clean video that I want to keep detail high on, I would rather use MainConcept or TMPGenc. They both produce video very close to original DV.

Here are some pretty good examples:

Procoder test @ 8500 kb/s - CBR


CCE test @ 8500 kb/s - CBR


MainConcept @ 8500 kb/s - CBR


TMPGenc @ 8500 kb/s - CBR


You can see in the Procoder test that the details are bured a little. In the CCE test, noise is introduced. In both TMPGenc and MainConcept test, the details were preserved the best.

[grill1968]

Yup, I too have being on the same belief for a while now as well. It is definately filtering. But, I would have to say, a damn good
one at that...

I totally aggree with you. While adding no or very minute extra noise Procoder even at 4000 bitrate kept the same or nearly the same details as other encoders did without filtering. Let's have a close look at the children's head and shirt on the I-frame 812 of my tested footage (see the encoder settings in my first post on the previous page).

DV original:

Procoder 1.5:

CCE 2.70:

TmgEnc 2.5:


dipstick: Could you please post the original DV pic?

manono: Playing the filter game in CCE did not bring the picture quality significantly closer to that of Procoder, although there surely was minor improvement achieved. I had to give up these filtering tests lacking enough time and knowledge. Perhaps someone could suggest "good" parameters which would fit to my test condition.

[dipstick]

grill1968 wrote:

dipstick: Could you please post the original DV pic?

Maybe you didn't understand what I did and maybe I should have explained. Each test includes both the original DV image and the encoded image. I simply made an animated GIF of both images, with each displaying for 1 sec. The idea is to stare into a section and watch for the differences as it changes from source to encoded versions. This is the best way to determine any differences.

I thought it was obvious, sorry if I confused anybody.

[grill1968]

I am so sorry, these gifs either did not work on my home PC properly or last night I was too tired to notice your excellent job. Thanks for that.

[grill1968]

dipstick: Sorry again, my firewall's ad blocking prevented your pictures being animated.

[Bourni]

I've been testing Procoder 2... I liked the resulting image quality but I must be doing something wrong because it's not fast as people says here. In a 5 minutes clip, I used CCE 1 pass VBR x Procoder2 1 pass VBR too and Procoder 2 took 15 minutes to do the job (0.42xreal time speed) against 3,5 minutes of CCE job. And I choosed High Speed option in Procoder 2. Why is this happening?

[BJ_M]

would need more info

[Bourni]

BJ_M,
I tested another 5:57 minutes clip of Constantine Avi movie (cutted in Vdubmod) in a faster computer (Athlon64 3200, 1G DDR dual channel):
Canopus Procoder2 Settings:
Video Biltrate 2500
Max Biltrate - 8500
Min Biltrate - 0
VBR 1 pass

Results:
Mastering Quality: 12:23
Highest Quality: 11:13
High Quality: 09:19
High Speed: 09:40

Then I used the same source file without audio just to compare encoding times:
High Speed - 09:06

As you can notice there's no significative difference in the encoding time between Mastering/Highest Quality, High quality/High speed and source file with or without audio...

With the same source file without audio and with same settings (including biltrate), it took just 2:20 minutes(1 PASS VBR PLUS 1 PASS VAF = 2 PASSES) to finish the encoding job in CCE SP 2,67.

Just to make another comparison, I encoded a 2h and 11 minutes without audio avi movie in Procoder 2. It took 4h and 37 minutes to do it in High Speed option. Then I encoded the same movie, but this time with audio in WinAvi to create the Vob files (medium quality). I know the final quality of a encoded movie in WinAvi is not good enough compared to other good encoders... But it did the job in 47 minutes!!!

So guys please tell me why you keep on saying that Procoder 2 is a fast enconder??? Good I agree, but fast???

[BJ_M]

it is fast -- and as reported on the canopus forums ...

1. upgrade to the latestest build - you need to have a registered copy to do so ..

2. NO 'cracked' versions work correctly, not saying you have one.... but many versions have some issues ..

3. it runs faster on some intel machines than amd (at least this used to be the case - it also runs slower on winXP64 (as do many other encoders))



it is is true that cce is faster - but procoder can do many things cce can't either ... also to effect speed is a lot of semi hidden settings in procoder (as in cce also to be fair)

[Bourni]

BJ_M thanks for your reply. I was not fair with CCE results. I tested an avi file.. I should had tested an AVISynth script. The CCE results went up to 04:55 minutes. Anyway I liked Procoder 2. I'm gonna use it sometimes. When I'm not in a hurry

[DVDEncoder]

Hi everyone... I just found this thread and I see that it goes all the way back to Aug 27, 2004, and is interesting that it is still alive. My question is: I want to encode from DV material to DVD and I concluded according all the answers in this thread that Procoder 1.5 is the best for dv to dvd. But in 2005 Canopus released a mayor update for Procoder 2.0 (It's service pack 1, ver 2.04.02) and it is not stated in this thread if they improved the quality of the mpeg encoder to match the one from 1.5. If fact the original test was prior to the update. Does anyone know if they did updated the mpeg encoder? Or just added other features. Thanks in advanced.

[BJ_M]

it was updated -- but not to 1.5 of course .. but improvemens were made to it

[Questmark]

This thread is great.

In regards to the old original comparisons-- I'm rather surprised how well Vegas 5 looked. I have Vegas 5, but have never used it to encode mpegs, and have had to move the files over to tmpgenc. What's the real difference in quality between the two of them as encoders? I saw the pics, but it's hard for me to see what I should be looking for when comparing them....

If I go to other boards where Vegas is the main topic, well, they all encode with Vegas to mpegs, rather than the encoders you've all mentioned here. Are they producing noticeably sub-par encodes?

Just looking for some info on this.... as it would be nice to simply encode in Vegas if I'm editing a project.

[BJ_M]

vegas (more so with version 6) uses a highly tweaked version of the main concept encoder .... it is not bad at all with dv sources and does a excellent job for HD -- it is not so good at re-encoding mpeg sources

[InXess]

"Highly tweaked"... could you explain in what way? Can MC 1.4 or 1.5 be tweaked the same way? I doubt that they have rewritten the engine... why would they give Sony a better encoder and keep crap to themselves...? NLE makers just keep on coming to MC even though their encoder sucks... and ask them to tweak this piece of junk so that it could drive their Pro grade NLE like Premiere...? I don't think so. I just don't get the logic here. MC is incorporated in many NLE's, in fact they have almost monopolized licensed encoder market (Sony, Adobe, Ulead and probably few more).

Do you have any details on those tweaks? It seems to be slower then stock MC Enc... and locked in a higher, more elaborate (time consuming) encoding scheme but that could be accomplished with MC too.

[BJ_M]

i have some details - yes ..

but I am not at liberty to say what they are ...

[FulciLives]

i have some details - yes ..

but I am not at liberty to say what they are ...

LOL

- John "FulciLives" Coleman