MPEG2 Encoder test

[Edmund Blackadder]

ADDED:
(in my samples post above)

ProCoder 2.0:
2 pass VBR, Mastering Quality, DC Precision 10.
Both, quiet and windy shots were included.

As you can see ProCoder 2.0 is doing a pretty darn good job and still outperforms CCE and TMPGEnc. So for those of you who are thinking of getting ProCoder Express or ProCoder 2.0, you are still going to get the results that are superior to CCE or TMPGEnc on interlaced DV footage. ProCoder 1.5 does have an edge (with the backgrounds images and also with ultra low bitrates), but version 2.0 is still very very good and will give you much more than you used to get from CCE or TMPGEnc.

[adam]

Edmund Blackadder I'm suspecting that maybe you didn't control the picture types of your samples. Unless you make them all B frames, or all P frames, etc... like BJ_M did, then you are comparing apples to oranges. Also it would be a good idea to expand your luminence range of your source sample (not the actual DV footage, but the still pic you take) to 0-255. All of your encoded samples are 0-255 and your DV is 16-235. That's why there is a visible difference in luminence, and honestly, why its difficult to really draw any comparisons. I'd also suggest biting the bullet on filesize and using .bmp or .tiff. JPG is far too lossy to assume that all your sources compressed the same. Ultimately, I think still pictures really don't mean anything at all, though I know BJ_M would disagree.

I think Procoder 1.5 is the absolute best encoder for pure interlaced footage as well, but it is hardly a matter of fact. Many will prefer CCE's sharpness and preservation of detail to Procoder's slightly more blurred image. I know I certainly do for other progressive sources.

Also, CCE is not an easy encoder to use. I suspect that if you played with it more you'd get better quality. For instance, there is no general sweet spot for the IQP setting. This is something that must be adjusted according to your source and your target bitrate.

[cosmin]

Here's another question about ProCoder 2: is there (usually) a big difference between the Highest Quality and Mastering Quality settings? I'm asking this, because the affordable ProCoder Express does not offer Mastering Quality.

[BJ_M]

sometimes yes - sometimes no ...

many times i see no diff. or speed diff.. i have not been able to figure out why ..

[Edmund Blackadder]

Edmund Blackadder I'm suspecting that maybe you didn't control the picture types of your samples. Unless you make them all B frames, or all P frames, etc... like BJ_M did, then you are comparing apples to oranges.

You're right about the fact that I did not pick any specific B, P or I frames. But I did pick the exact same frames of the clip. And if one of the encoders shows at least one really crappy frame and the rest of the bunch are alright, the crappy one will still be visible on your TV in full speed. The reason I picked those particular frames is because they represent really well what I was seeing on my TV in full motion, precisely CCE's picture being very dirty and TMPGEnc being to blocky. Adam, if you would like to see the MPEG2 clips or DV and compare for yourself I'm more than happy to share them with you. Just let me know where to send/upload them. Perhaps you have your own FTP or something where I can leave the clips?

Also it would be a good idea to expand your luminence range of your source sample (not the actual DV footage, but the still pic you take) to 0-255. All of your encoded samples are 0-255 and your DV is 16-235. That's why there is a visible difference in luminence, and honestly, why its difficult to really draw any comparisons. I'd also suggest biting the bullet on filesize and using .bmp or .tiff. JPG is far too lossy to assume that all your sources compressed the same.

I can expand my DV frame to 0-255 if you wish, but I didn't want to change anything in those stills, just to avoid any potential flaming for biasing testing results. Though, it's not that difficult to draw the comparicons. After all the DV frame was there just for the general reference. The main comparison was not even about the luminance or colors, but about the artifacts. And I think I got my point across when it came to artifacts. On a side note, the JPEG's were made with side by side comparisons with the original TARGA captures in Adobe ImageReady with Maximum 80 quality and are indistinguishable from the originals. Even BJ_M's sample JPEG's seem to be more compressed than mine, according to file sizes (minus PAL/NTSC resolution differences). Videohelp's limit to posted images is 150kb per file and I came close to that limit with these JPEG's. So don't worry about these JPEG's giving you false artifacts, they don't, please let me assure you that.

I think Procoder 1.5 is the absolute best encoder for pure interlaced footage as well, but it is hardly a matter of fact. Many will prefer CCE's sharpness and preservation of detail to Procoder's slightly more blurred image. I know I certainly do for other progressive sources.

Well, according to some people on this forum, and I agree with this idea, CCE's perceived sharpness is only due to it adding noise. That will always make things seem sharper. And in the end, why then that leaf sample looks so much less detailed in CCE than in ProCoder? I know, I know, B, P and I frames, but still it doesn't look well preserved and is in fact more blurry than ProCoder's result.

Also, CCE is not an easy encoder to use. I suspect that if you played with it more you'd get better quality. For instance, there is no general sweet spot for the IQP setting. This is something that must be adjusted according to your source and your target bitrate.

Honestly, I gave up on Cinema Craft for my DV footage. I played with many parameters on my own, then read the manual countless amounts of times trying to find some advice for good parameters according to Custom Technology and still it isn't good enough. Yes, I could be playing with CCE for various kinds of footage and bitrate, but that would still be mostly a waste of time and would not give me satisfaction, while ProCoder does it perfectly with the default settings. All you have to worry about in ProCoder is Progressive/Upper/Lower Fields First and bitrate selection. Everything else is nearly perfectly tuned by Canopus. Nowadays I don't even use CCE for the progressive non-compressed-AVI originated motion menus that I do for my DVD's. ProCoder looks just as good with them as CCE. So, as of now, I see absolutely no reason to use neither CCE nor TMPGEnc for any of my encodes (alright - 99% of them), be that progressive uncompressed graphics or interlaced DV footage.

[Edmund Blackadder]

I just noticed this:

Upload image or file(Optional)
Type filename.zip in your post where the file-link should be. Max file size is 2 MB.

Does that mean that I can chop my test MPEG2 clips down and upload a few 2MB RARed segments so people can download them all and later reconnect in WinRAR? Or is that gonna be pushing it and I'll get banned for doing so?

Anybody knows?

P.S.: There's about 16MB worth of these leaves'n'buds MPEG2 clips.

[BJ_M]

dont right now -- pushing a lot of bandwidth ....

limit it to 2 x 2 mb at the most ...

or host them on your own server

[BJ_M]

plus - though appreciating the nice test images - the content is not a real test of encoders as the motion vector dual prime predictions are little changed from frame to frame per macroblock.

they ARE a good representation of many of the factors you have pointed out though ---

[adam]

Edmund Blackadder its impossible to compare encoders if you aren't matching picture types. Comparing an I frame to a P frame is like comparing a 5Mbit encode to a 2Mbit encode.

I wouldn't necessarily expect CCE or TMPGenc to fare any better in a more controlled test with this source. For DV sources, Procoder is my baby too. But Procoder is hardly the be all and end all of encoders. There are several other encoders which provide equal or better quality depending on the source types, settings, and preferences of the viewer. This is all far too subjective to ever declare a unanimous winner.

Myself, I use different encoders for different jobs.

[cosmin]

It would be so nice if someone could post a test with MainConcept. Apparently, big players (Adobe, Sony, Panasonic) seem to love it, although it received less than favorable results at videohelp, when compared to ProCoder and CCE. I wonder how this one would handle the leafs scene currently discussed in here.

[Edmund Blackadder]

It would be so nice if someone could post a test with MainConcept. Apparently, big players (Adobe, Sony, Panasonic) seem to love it, although it received less than favorable results at videohelp, when compared to ProCoder and CCE. I wonder how this one would handle the leafs scene currently discussed in here.

Cosmin, currently I don't have an access to MainConcept encoder. But from what I remember when I tested it awhile back (the standalone version), there was a little bit of both - noise and blocks. I liked it better than TMPGEnc and I think it even looked better than CCE for my DV tests.

The output from Adobe Premiere through built-in MainConcept encoder wasn't very good.

But what surprised me the most was Discreet Cleaner XL 6 output, again using the built-in MainConcept MPEG2 codec. It actually looked very good. I guess it's because Cleaner used some sort of its own very effective filtering. But that was awhile back.

Edmund Blackadder its impossible to compare encoders if you aren't matching picture types. Comparing an I frame to a P frame is like comparing a 5Mbit encode to a 2Mbit encode.

Adam, I could use that 2x2MB space that BJ_M said I can use, and RAR the 1 full second of captures of each encoder into medium quality JPEG's, just so you see all 25 frames from each encoder and pick whichever frame you like the most. But I don't think that would change much of the overall results.

There are several other encoders which provide equal or better quality depending on the source types, settings, and preferences of the viewer.

I'm not saying that ProCoder is the best of all encoders. I'm absolutely positive that the majority of high-end hardware encoders used in Hollywood will beat ProCoder any minute. But talking about the encoders that are within a reach of us, regular mortals, ProCoder is... Well you know my opinion.

At least we both seem to agree that ProCoder is best out of the popular mainstream encoders for interlaced DV footage .

[J. Baker]

Why doesn't any body test YMPG? When I was doing my own test a while back, it seemed YMPG did the best on detail and still keeping the video pretty close to the original but I'm still new to this and would like your thoughts.

[BJ_M]

It would be so nice if someone could post a test with MainConcept. Apparently, big players (Adobe, Sony, Panasonic) seem to love it, although it received less than favorable results at videohelp, when compared to ProCoder and CCE. I wonder how this one would handle the leafs scene currently discussed in here.

go to page one of this thread -- i did test it

[canadateck]

MainConcept does not look to good ,And vegas Looked better.I dont get it.

[BJ_M]

sony/sonic foundry tweaked the hell out of the encoder

[FredThompson]

Edmund Blackadder, I think you need to lay off the insults...it's only showing how immature you can be.

Listen State Of Mind. I wasn't the one who started the pointless insults in this thread. Everything was nice and dandy in a good spirit of discussion until Fred T came here and started spilling bullshit all over what I had to say with absolutely no backing to his words. So if someone is immature in here is Fred T for hijacking a peaceful thread and turning it into a big argument.

Wrong, again.

My first post quoted you directly claiming PAL is better "everything" than NTSC. The proof was your own post. It was your choice to behave as a boorish troll while spreading false information. I find you to be a waste of time and potentially quite injurious to other people.

[FredThompson]

Disclaimer: This post is directed towards people who are truly interested in the relative strengths/weaknesses of PAL DV vs. NTSC DV.

Neither PAL DV nor NTSC DV fully encapsulate the qualities of the other. Therefore, it is impossible for one to be superior in all manner to the other. PAL DV has an smaller vertical chroma resoultion and a lower frame rate than NTSC DV.

For those who are truly interested in MPEG2 compression of NTSC DV source, I highly suggest you use the AviSynth filter Reinterpolate411. NTSC DV is 4:1:1 which is one chroma sample shared for a horizontal grouping of 4 pixels. PAL DV and MPEG2 also share one chroma value for a grouping of 4 pixels but a square group, not a rectangular one. There is no 1:1 correlation between the models so some adjustments must be made. This is why the Reinterpolate411 filter was developed.

PAL DV and MPEG2 have their own weaknesses. Red is a mess. Put simply, this is why the Reinterpolate420 filter was developed.

For more information on these AviSynth filters, search at doom9 or neuron2's board.

[BJ_M]

CCE 2.70 does this conversion (when checked(new feature) ) , so does vegas (as it always feeds RGB internally) , procoder has its own conversions as well, but diff. ..

[lordsmurf]

Why doesn't any body test YMPG? When I was doing my own test a while back, it seemed YMPG did the best on detail and still keeping the video pretty close to the original but I'm still new to this and would like your thoughts.

Crap. For many reasons. It's a lot like BBMPEG or something else from about 5-6 years ago. It's to be expected from freeware/cheapware encoders.

[FredThompson]

CCE 2.70 does this conversion (when checked(new feature) ) , so does vegas (as it always feeds RGB internally) , procoder has its own conversions as well, but diff. ..

Are you referring to my comments about colorspace? If so, I wonder if the folks at CinemaCraft were watching the developments in the AviSynth realm.

Direct conversion to RGB, alone, won't do anything to "fix" the chroma issues. The artifacts can be made less obvious using interpolations.

I've used Vegas in the past but avoid it as much as possible. RPG makes sense for initial generation of CG but it really screws up non-RGB source. RGB and YUV don't have the same range of values so converting between one or the other is always lossy. I'm sure you know that, I'm just explaining to readers who don't.

Here are some links:

Reinterpolate411: http://forum.doom9.org/showthread.php?s=&threadid=58294
Reinterpolate420: http://forum.doom9.org/showthread.php?s=2d5fe189d4001190388d481177b0928e&threadid=82787

411Helper: http://www.geocities.com/xesdeeni2001/ (VirtualDub method)

4:2:0 Chroma Upsampling Bug: http://www.hometheaterhifi.com/volume_8_2/dvd-benchmark-special-report-chroma-bug-4-2001.html (thanks, Ivo)

[cosmin]

There is no 1:1 correlation between the models so some adjustments must be made.

In spite that you're bashing 4:2:0 chroma subsampling used in PAL-DV, both PAL-MPEG2 and NTSC-MPEG2 use 4:2:0. What is less obvious, but still important, is that both PAL-DV 4:2:0 and PAL-MPEG2 4:2:0 have the chroma positioning co-sited (unlike PAL-MPEG1), which means that the YUV samples decoded from PAL-DV get translated to PAL-MPEG2 without any modification whatsoever. So yes, there is a 1:1 correlation between PAL-DV and PAL-MPEG2. And, no, you can get no better than this. You can get equally good results with 4:2:2 to 4:2:0, and also with 4:4:4 to 4:2:0, but you can never get better results. Reinterpolate420 is useful when displaying PAL-DV material, but not when converting PAL-DV to PAL-MPEG2.

But NTSC-DV chroma subsampling is neither 4:2:0, nor 4:2:2, nor 4:4:4. NTSC-DV is 4:1:1, and that's not good when trying to transfer it to DVD.

And in spite that you're praising NTSC's 4:1:1, neither your initial attitude nor your belated points about the Reinterpolate411 filter can change the mathematical fact that there is loss when going from 4:1:1 to 4:2:0. In theory, slightly more than half of the information is lost when using a good DV codec (or a dumb DV codec helped by Reinterpolate411), and significantly more than half of the information is lost when using a dumb DV codec without Reinterpolate411. ("Dumb DV codec" means one of the old Canopus or MainConcept codecs that were too lazy to interpolate the chroma themselves.) It is worth noticing that there are only dumb NTSC-DV codecs, but there are no (and there cannot be) dumb PAL-DV codecs :P

In the end, you get a DVD with 4:2:0 MPEG2. No matter if it's PAL or NTSC, applying Reinterpolate420 is equally useful (or equally not useful) to both of them.

And in spite that you're arguing vehemently for NTSC's 60fps rate and against PAL's 50fps rate, this detail has no relevance to our discussion. This was said already. Please start a new discussion thread about this, if you're so interested in it.

Disclaimer: This post is directed towards people who are truly interested in the relative strengths/weaknesses of PAL DV vs. NTSC DV.

Your disclaimer is not good enough to excuse you for what you're doing. If your post were truly intended towards people who are truly interested in whatever you're saying, you should have started a new discussion with an appropriate subject title - and not hijack this one once again!

This discussion is already long enough and hard to read. It is full of very useful information, thanks mostly to BJ_M and Edmund Blackadder who did not just express simple opinions, but did spend the extra effort to provide extremely detailed visual and technical argumentations. Too bad that the discussion got polluted with noisy, irrelevant quarrels started by heavy AND unbacked statements coming from you.

The proof was your own post. It was your choice to behave as a boorish troll while spreading false information. I find you to be a waste of time and potentially quite injurious to other people.

It is a fact that you had posted unbacked claims. (Valid or not, they were unbacked.) It is also a fact that you had been the first to throw insults. It is also a fact that your unbacked insults had been targeted towards a poster who had gone through the effort of providing extremely detailed info to back his claims. I acknowledge the fact that you are coming with factual information in your latest post, but (1) you should have done this in the first place instead of giving us nothing useful but attitude, and (2) your belated arguments are flawed. If you disagree, feel free to start a new discussion about the value of 4:1:1 chroma sampling and the 29.97fps rate, the weaknesses of PAL, and the merits and the demerits of Reinterpolate411 and Reinterpolate420. Invite me there with a PM, and I'll tell you where you're right and where you're wrong, and I'll give you all the backing I have (and you can also give me there all the backing that you have). But for crying out loud, Please Leave this discussion about MPEG2 encoder tests alone, if you have nothing relevant to say about MPEG2 encoder tests

[FredThompson]

[comment purged by author]

edit: "Outrage" regarding comments erroneously attributed to me is not my responsibility.

Colorspace conversion issues are important and completely relevant. The initial discussion made no distinction between PAL or NTSC DV as source. As such, analysis and comparison is polluted unless this is discussed. People who might attempt to replicate test results most certainly need to start with the same conditions and understanding.

Edmund Blackadder made the erroneous claim that PAL DV is superior, in every way, to NTSC DV. This simply isn't true. Left unchallenged, they serve to confuse people who don't know colorspace issues.

[Comment purged by author]

edit: frame rate has relevance to encoding because of its relationship to motion. Remember, encoding involves analysis of change between video frames. Also, I'd like to take this opportunity to point out (for "newbies) that PAL (NTSC) interlaced source is not really 25 (~30) fps, it's actually 50 (~60) fps made of 2 "woven together" sets of sequential images. It is quite common to see discussions which mention deinterlacing source. If source is interlaced video, as opposed to combed film, the most accurate representation of the original is to leave it in the original interlaced configuration. Deinterlacing video source will really mess up the motion analysis because it's trying to "average" together motion of 2 different times. If the motion isn't perfectly linear, the "average" leads to corruption.

I DO understand what you're saying about the pure mathematical aspect of lossy encoding of data. Looking back at it, I was thinking about how quality of encode can vary dramatically due to bonehead data prep. Point taken.

[cosmin]

But for crying out loud, Please Leave this discussion about MPEG2 encoder tests alone, if you have nothing relevant to say about MPEG2 encoder tests

I do acknowledge that you did post something relevant about the subject. Please allow me to apologize for my rushed and hard statement.

[FredThompson]

Apology accepted, as are your points about relevance to the thread topic.

[Edmund Blackadder]

Thank you Cosmin, you're my hero!

And now, comrades, I will take a vacation from posting in this thread. That will hopefully make things settle down a bit. I will still monitor this topic though, just to see if anyone gets any more tests done. Perhaps I'll go and revive my ancient "DVD+R9 Double Layer first burn: problems with layer switch" thread that got quite a bit of attention and I seemed to be doing better in there, without any conflicts. By the way if you are at all considering making your own highly compatible DVD+R DL discs from scratch - after converting to MPEG2 with such wonderful encoders that we discussed here - I definitely recommend you reading that whole thread.

Cheers!

EB.

[Skynet107]

I never thought I was see a fanboy of the encoder world but Edmund Blackadder proved me wrong.

Like it was stated several times, no encoder is the end all be all for all types of material.

[vhelp]

Sorry for not responding guys.., I lost this thread. Anyways.

@ cosmin

At this point in time, I know NO codec that handles 411 bug built-in. I've
never seen it, else I'd be using it
One just have to use a 411 filter (or variation of it) somewhere's in your
source before you encode it (Sometimes, I just leave it alone and make do w/
what I got, and move on. Just depends on my moods) Anyways.

I will make a VIDEO_TS out of it and I can still try to post it. Do you wish to see it, Vhelp? Though it's still best if you have a PAL capable TV.

Yes. I'd be interested in seeing some samples of your Procoder samples
.
I'd appreciate it if you can find a place to post this for all us curious
peoples to D/L easily. I don't fiddle w/ e-mail and stuff. Sorry.

On the subejct of Encoders ...

I've spent the good part of two weeks w/ Procoder vs. TMPGenc (and my usual
methods of encoding techniques) and I did find a plus with Procoder
I'm with Adam and Edmund w/ Interlace sources (ie, DV footage) - the quality
is very good. I've figured out how to "operate" PC 1.5 w/ DV footage source
IMO, it's fantastic. Also, (in addition to my basing this w/ mine own eyes )
I also based this on a commercial DVD (I still have "ripped" pieces on my hd) of
Star Trek "Jean-Luc Picard Collection" set. And, the quality from my DV encodes
w/ PC 1.5 were virtually the same or in resemblance w/ the (how did Edmund put
it in an earlier post.. fluid or something) with the DVD and my DV encode.
.
I know that no one has given any thumbs up on Film sources encodes w/ PC 1.5
but I would be curious to know what some of you who do use it, have to say
on its quality. I've done some myself, and have comments to say, but much
rather hear those that have used it with it first.
.
I do have a more elaborate platform for dealing (processing) video w/ PC.
for film source materials..

.. For Film sources:
.. DV -> vdub [filters] -> tmpg [filters] -> Procoder = next to TMPGenc
.. results!! ( <-- edited was too quick w/ cut/paste )

And, there's a method to my madness. Really. There is.

The 411 Filter for DV ...

The 411 filter can be found in (at least) this version of vdub v1.5.10 built-in.
Look under Video\Filters\Add..\chroma smoother and it has several
setting, depending on your INPUT source. And I use this w/ great results.
(Note, I don't like using filtering, but sometimes, you just gotta)

And now, I think that maybe we can narrow down the Encoders and their versions,
and begin a new level of sampling some video perhaps ??

* TMPGenc
* CCE
* Procoder v1.5

I hope that more test samples can be posted here soon, for our testing. This
thread has ben very interesting so far, and I continue to look forward to
more comments/tests/theories/analisys and other fun stuff.

Thanks to all those who contribute.

From the Video Workstation of,
-vhelp 3012

[FulciLives]

I thought TMPGEnc Plus could account for the DV 4:1:1 sampling rate if you are using the Canopus DV codec.

There are several options built-in just that that DV codec i.e., the Canopus codec.

- John "FulciLives" Coleman

[vhelp]

..regarding 411 filtering..

Yes. I believe it so. (or so I thought) I was using the canaopos DV codec
( after creating a "cdvc" ) avi, all options in TMPG should be turned off, w/
the exception of the [x] Interpolate YUV data from 4:1:1 to 4:4:4
w/ this "cdvc" avi and it should do the same as a 411 filter, or help it some.
Last time I used it, (few weeks ago) it worked. I guess it depends on the
source and how you "feed" or frameserve it into TMPG, and weather or not
you have the other param settings ticked or not, [ ] Output YUV data as..
.
.. will also depend on your source.

-vhelp 3013

[jdw]

Thanks, Edmund Blackadder, for the tip about ProCoder 1.5 being better than 2.0. I find this info very useful.

There is a problem, though: ProCoder Express uses the ProCoder 2.0 engine. There is no "affordable" version of ProCoder 1.5, right? Or, is ProCoder 1.5 still being sold, at all?

For those who have Premiere Pro and is looking for PC 1.5....1.5's built in framserver plugin does not work. You can however frameserve to PC 1.5 from PPro using Debugmode FrameServer. PC 2.0's frameserver plugin does work.

Also, I could not get PC 1.5 to recognize AviSynth AVS scripts, where as PC 2.0 does recognize them.