MPEG2 Encoder test

[funkguy4]

Hey guys, I just recently worked w/ some DV footage w/ editting, and I must say, CCE does not like unclean/curvy source. I had this nice wavy font for a title that CCE encoded perfectly while the title scrolled, but the moment that the title froze in place a while, CCE gave me a very gridlike image, even w/ the highest of quality settings. I was confounded, I checked the frameserver AVI, I tried to encode w/o deinterlacing, to no avail! I had to bust out the trusty TMPGEnc, which gave, while not a perfect picture @ the still scene, a picture MUCH CLOSER to the source and less staticy as well. I guess TMPGEnc is my DV encodwhore, and CCE my transcodwhore (w/ DVDRebuilder )

[hiptune]

I would like to throw in an analagy, if I may, to explain why I continue to use CCE in most cases.

<snip> but the difference in output in no way justifies the difference in times. We are talking an order of degrees difference in quality versus orders of magnitude difference in times. I cannot afford that luxury. I settle for very good, instead of a little better than very good. I am happy with my choice. I will continue to watch and experiment, as we all should, because encoders constantly change. But until the diferences (quality, speed) have much less of a gap, I will stay where I am.

Nice of you to drop in and give your thoughts on the matter of very slight (to you) differences.

But myself and many others here have a job to turn in. Some of these projects will be displayed on very large screens. And that very slight difference in sharpness of lack of blocks in the grain, can make you a star in your audience's eyes.

I just did a demo reel for a Director of Cinematography or "DP" for short. This guy has worked with Tim Burton and other A list directors. This is a guy who has shot major Hollywood movies. He knows what his pictures should look like. And he took my DVD (encoded with Procoder) home and watched it on a very big screen. He called that night to say he loved it!

I didn't encode with "very good" setting believe me, Mastering all the way!!!! I need these types of clients, and I don't care if it takes a week to encode something, my disc has to look absolutely stunning and play in every machine. My future depends on it.

There are others too here who need that little tiny extra edge for their projects. And that is what this thread is about.

[The_Doman]

I personally love the procoder results too.

On a german site they did a blind-test with the popular encoders.
Procoder was indeed the most liked.

http://www.gleitz.org/index.php?id=97

[BJ_M]

thanks for the link -- read it in english here

http://translate.google.com/translate?u=http%3A%2F%2Fwww.gleitz.org%2Findex.php%3Fid%3...language_tools

[dphirschler]

Why a minumum bitrate of 1500? Why not zero?

Good to see TMPGenc holding its own. I've been satisfied with that encoder for a long time. It takes its share of beatings for being slow and only offering two-pass VBR, but the quality looks pretty darn good to me.


Darryl

[vhelp]

Encoders as tools ... (Procoder vs. TMPGenc) ...

I also think that you have to look at Encoders as tools (in another way)
ie, "per project type" vs. using that all-in-one Encoder. I love TMPG. I
always will. But, I think that for pure Interlace sources (I'm sticking to
my DV home footage here) Procoder seems to give the best results - every time
(pending your settings of course)
.
The one thing that I lacked (still lack) in TMPG's area of perfection with
video sources, is those sources that are Pure Interlace. I haven't spent
much time with debuggin it over the years, and perefecting, until recent. I am
now getting closer to nailing 'quality' in that source type too. Its just taking
some time to complete.
.
Also, and from my recent tests with TMPG vs. Procoders 'pure interlace' source
encoding, there is a slight difference, but both look very good. Its now just
a matter of: TV set; preference; the eye; and what one views as quality in a given
MPEG-2 on their end.
.
But, like I said, I'm still debuggin (fine-tuning) this area of TMPG.

You also have to factor in, those certain tweaks (not the encoding settings)
that a skillful user has in his/her bag of goodies. The EYE.., and the Instincts.

-->

Procoder is still limited in certain aspects. ie, in TMPG, I can set an IVTC
a certain way, (pending source) and it will always match the pattern perfectly
for a super encoded MPEG-2 final project. As much as I've tossed and turned
with Procoders' IVTC mechanism, it doesn't hold the same candle as TMPG's.

Some day, maybe I'll post a 'mpeg encoding test' too
Actually, I've ben meaning do this a very long time, but I keep myself berried
inside countless projects, that I don't have the time or energy to begin it.

-->

But, for now, I'm continuing with TMPG for both Film and Interlace contents,
and will stick those special Procoder projects.

-vhelp 3117

[hrlslcbr]

Well, it seems that this thread is dead, but anyway....
BJ_M: I know you have CCE SP 2.7; could you encode the same clip with it to see if the new encoding engine is better?
Because when I downloaded the trial, I saw a BIG speed increase, but I want to know if it produces better quality too.

Thank you.

[gwoiler]

This seems to be a good place to ask this:

When using open GOP or closed GOP with TMPGenc plus.... is one or the other non-DVD compliant? I get some wierd problems with DVD's if they have some still video and a chapter happens to fall within this still area. WHen the motion starts up it kind of stuters while the audio marches on. And then the video catches up to the audio and all is well. Sure would like to figure out what I am setting wrong. Bitrates are usually above 4800. CBR and VBR seems the same.

Glenn

[adam]

Both open or closed, or a combination of open and closed GOPS ar perfectly compliant. Generally you always want to use open GOPs (GOPS will still be closed at scene changes) but it seems there is a bug in TMPGenc that results in incorrect bitrate allocation if you leave gops open.

In any case, I don't believe this could be causing your problem though it could be that your gops are too long.

[kinglerch]

I recently did a project with Procoder and TMPGEnc. I didn't want any additional filtering or adjustments, I spent days with my AVIs to be sure they were "just right".

When I sent the AVI into Procoder 1.0 (mastering 8000 2 pass VBR) the image was smashed, far too dark, poor colors...I figure it is the older revision causing the problem.

I sent the same AVI into Procoder 1.5 (mastering 8000 2 pass VBR) the image was much better than in 1.0 but it was still darker than my source and some subtitly was lost. This may be good for some sources, I don't know. Using the 601 expand/shrink either made it too bright or even darker.

TMPGEnc (8000 CBR) did not darken or lighten the picture and looked most like the source, which was what I was trying for. Your results may vary, I did not check blockyness, bitrate, etc. Only that the result looked most like the source I worked so hard to get correct. TMPGEnc created the best result in that regard.

EDIT: I examined further Procoder with the 601 shrink and this worked much better for my source. The brightness was just as I expected. Sorry for the misinformation. For anyone experiencing brightness/contrast issues with Procoder, try the 601 shrink or expand.

[troyvcd1]

Procoder alters a videos colors. Some people like this though.

[BJ_M]

Procoder alters a videos colors. Some people like this though.

this was shown in tests - with smpte test charts, to be not the case - at least with the type of source files i used ..


scroll back through the pages to see the results

[kinglerch]

Procoder didn't change my colors, mainly the black levels. But when I set the 601 to shrink, the result was the same as the source color and blacklevel.

What's with the spider for your icon and sig? It's distracting and creepy so I have to scroll past it to read your posts.

[grill1968]

First of all, thank you for all your contributions in this topic, I have learned a lot. As a newbie I don't know if it's worth of warming up this stuff but recently I have faced some problems which others may be also interested in. I was curious whether choosing different mainstream encoders (CCE, Procoder and TmpgEnc in my case) does really matter regarding picture quality of non-professional (e.g. amateurish) DV footage recorded with a low-price camcorder? In addition, I wanted to test how the interlaced material looks on a relatively large (19' actually) TFT monitor at full screen. For certain reasons I was also interested in as low bitrate as 4000 avg. In this test I was concentrating on the noise and blocks, which mainly annoy me when I'm watching video.

N.B.: Below it is my personal opinion. I'm not to make judgement that X encoder better than Y, but I can honestly say I see X better than Y.

Source: 1 min interlace DV avi recorded with hands in a half-cloudy day, pans/scans of people feeding pigeons in a square (a tipical tourist's theme).

All encoding was done at interlaced (BFF), 4000 avg VBR, 1500 min, 9000 max, 10 DC. I set up the number of VBR passes and/or quality settings so that the encoding time would be similar (ca. 6 mins actually) for each encoder. Other settings:

CCE 2.67: 1+5 pass VBR, lum 0-255, quantizer 27

CCE 2.70: same as CCE 2.67 + 30 V/C

Procoder 1.5: 1+1 pass VBR, mastering mode.

TmpgEnc Plus 2.5: 1+1 pass VBR, Motion search precision: High Quality

Since I'm not intending to share lossy compressed sample images with you, so please let me briefly give you my impressions without showing images. Anyway, I can send these images (ca. 5 Mb) to anybody being interested.

Results:

My ratings form 1 to 5. 1: very bad, 5: very good:
(noise/blocks/overall performance)

Entire video:
CCE 2.67: (3/4/3)
CCE 2.70: (3/4/3)
PC 1.5: (5/5/5)
TmpgEnc 2.5: (4/3/4)

Sample images [two I-frames (692, 812) from each mpeg2]:
CCE 2.67: (3,4,4)
CCE 2.70: (3,4,4)
PC 1.5: (5,5,5)
TmpgEnc 2.5: (4,3,3)


Briefly, CCE's were much more noisy, TmpgEnc was much more blocky than Procoder. The two CCE's were almost the same. Interestingly, I feel TmpEnc's I-frame quility worse than that of CCE's in spite of that I liked TmpgEnc better while I was watching the video. Perhaps the reason could be that the blocky background annoyed me much more in the still pictures as compared to the movie. I saw the I-frame picture quality of Procoder very close to the original. The video quality of Procoder is still far from the perfection but it did, though, the best (and acceptable) performance for me in such a heavy condition (low bitrate, full-screen on TFT) . However, my eyes are absolutely not so trained as those of others working on the film, photo etc. business .

Any advice, comments, other test results will be happily accepted. I'm very interested in fine-tuning the tested encoders and how the others work in similar condition.

P.S.: Sorry for my English

[troyvcd1]

I have always found on first look that ya procoder looks very good with low noise and bright vibrant colors. But you really need to look at skin tones to see "to me" that procoder alters the contrast and produces unnatural pictures. No need to come back and blast me about procoder not altering the contrast and what I need to do to stop this. I have tried all the suggestions and still get the same results.

[grill1968]

I have always found on first look that ya procoder looks very good with low noise and bright vibrant colors. But you really need to look at skin tones to see "to me" that procoder alters the contrast and produces unnatural pictures.

Thanks for sharing your experiences. I will check the contrast out again.

[manono]

Briefly, CCE's were much more noisy,

Try using CCE with a better quantization matrix. A lot of the noise comes from the crappy default Standard Matrix that's used by CCE. Hint: get a decent matrix off of a good quality DVD and type it into the matrix boxes of CCE.

[grill1968]

Hint: get a decent matrix off of a good quality DVD and type it into the matrix boxes of CCE.

Thanks for the hint. How can I get the matrix off a DVD?

[manono]

Hi-

You can get the matrices used in any one of several ways:
1. Decrypt a small vob (like a chapter) from a good commercial DVD, open it in the latest version of DGIndex, go Options->Log Quant Matrices, and then File->Save Project. Whatever matrix or matrices used in that particular vob will be in the Quant file created.
2. Open a vob in ReStream (Files Of Type->All Files). Hold the mouse arrow over the 2 blackish boxes inside the Intra and Non-Intra boxes to see the Intra and Non-Intra matrices used.
3. Get the Rockas Rebuilder Matrix Editor:

http://dvd-rb.dvd2go.org//modules.php?name=Downloads&d_op=viewdownload&cid=24

It includes a number of matrices laid out for you to copy over into the CCE boxes and save. The first one, labeled as "1. High-High" (aka Fox home Entertainment) is used by Criterion, Fox, and Warner (they use a very similar one) and is just about the best matrix used by the best studios. Number 2. High-Medium is also quite good. Both are way better than anything available in CCE.

Now, I work mainly with DVD backups of movies, although these matrices are used with interlaced DVD video as well. But not having ever worked with DV footage before, I don't know for sure how well they'll work. I would guess they'll improve its appearance as well, and should certainly improve on the amount of noise. They'll also give you sharper and more detailed results, with a possible increase in artifacts, depending on the resulting average quants. They are relatively high bitrate matrices, much more so than is the default Standard Matrix of CCE. I don't know what matrices are used by Procoder and TMPGEnc. Maybe if you have time, you could follow the directions in "1." and post them here. I have an educated guess as to why everyone says the Procoder results for interlaced video are so much better, but I'd like to see the matrix, if possible, to confirm my suspicions. Thanks.

[grill1968]

manono:

Thanks for your valuable post. If I have enough time and chance to use the required tools I'll surely try playing with the matrices. In my test I applied the default/standard TmpgEnc/CCE matrices which are entirely the same by numbers. However, the blocks and noise produced by these 2 encoders were not the same (as I saw them, of course). Procoder 1.5 does not show its matrix.

I have always found on first look that ya procoder looks very good with low noise and bright vibrant colors. But you really need to look at skin tones to see "to me" that procoder alters the contrast and produces unnatural pictures.

Thanks for sharing your experiences. I will check the contrast out again.

In my test all encoders changed the colors and the contrast in a certain degree (at least on my non-professional TFT monitor Samsung SyncMaster 913v). Here's an illustrative example from the I-frame 692:

DV original:

[grill1968]

CCE 2.67:

[grill1968]

CCE 2.70:

[grill1968]

Procoder 1.5:

[grill1968]

TmpgEnc 2.5:


I still feel Procoder being the best but I may be prejudiced (though I never used Procoder before). What kind of settings/filters can help to correct these colors/contrast in this situation?

[grill1968]

[I still feel Procoder being the best

Or TmpgEnc (regarding colors)? Getting to be uncertain looking over and over...

[hrlslcbr]

grill1968: What colourspace does the video have?
I have CCE basic and I don't get that kind of problems with colours.

[grill1968]

hrlslcbr:

My avi source is 24 bit RGB, actually it came through UleadMediaStudio Pro 7. In the CCE user's guide I read CCE uses YUY2 internally, so in case of RGB source the RGB->YCbCr conversion is needed. Is this the reason of my colour change? Should I use a frame server for the conversion? Anyway, could I and should I capture DV to YUV colour space? I apologize for expanding this thread but as I said earlier I'm so rookie in this stuff, especially in the world of colours . If you feel me persona non grata I'll get out then...

[hrlslcbr]

That's the problem. You should set the Luminance option to 0-255 (In CCE basic it's under 'Advanced Video Settings', don't know where it is in CCE SP). This should prevent CCE from changing the range of colours.

[grill1968]

That could not be a problem at least solely. I followed the CCE FAQ's recommendation and I set the lum. level to 0-255 for this test.

[hrlslcbr]

DV is supposed to be in YUV format, but some programmes might convert to RGB to edit it. If you can (and want), could you use avisynth to frameserve the video to CCE, and put at the end of your script:
'ConvertToYUY2()'
Thank you.
BTW, what did you use to take the pics from the video? I remeber using VirtualDubMOD to compare different encoder's output colours and it showed different colours for all CCE encodes, but they were all the same when playing them back (from different encoders).