MPEG2 Encoder test

[offline]

I've never had ouput as bad as shown above with CCE and I use CCE all the time. I always
prefilter and set Q to 3, 3 pass and get a nice
result.

LordSmurf: As quick as you are to condem CCE, I
wonder if you ever prefilter any of your sources or
do you just dump the DV in. DV has issues anyway so I mostly use Huffy for superior results.

[lordsmurf]

Why would I filter DV video if it's already shot perfectly the way I want it? It's not like it's degraded VHS.

[FulciLives]

Why would I filter DV video if it's already shot perfectly the way I want it? It's not like it's degraded VHS.

Perhaps to apply a 4:1:1 filter.

- John "FulciLives" Coleman

[SatStorm]

To eliminate the things we don't see, but the encoders do see. That's why we have to filter. Because the things we don't see, may produce problems, like macroblocks. And I don't know about you, but I had more than once to deal with macroblocks, even with a CBR 9800kb/s from a DV source, using TMPGenc and procoder (demo 1.5). A slight filtering and that problem disapeard, with NO visual cost...
That's my opinion always. The opinion of the hobbiest, not the industrial users...

[adam]

Why would I filter DV video if it's already shot perfectly the way I want it?

My guess as to what Offline was specifically referring to was the nuances of one encoder that you may find preferable, which can be emulated in another encoder through filtering.

For example, you've made it clear you don't particularly like CCE's quality, and that you feel it adds too much noise to the picture. Perhaps you like Procoder 1.5 for DV better. (I do.) But Procoder, by default, does quite a bit of softening to the picture. You could do this very same thing in CCE by using either its built in noise reduction filters or by prefiltering your source. You'd probably get results somewhat more like Procoder, maybe you'd even prefer them to Procoder.

[State Of Mind]

I've come to like AviSynth and will be using that will all of my videos. Even when it comes to trimming the video. Plus, with AviSynth, the special effects can be frameserved in and out of Adobe Premiere. I am the pickiest guy on Earth when it comes to quality. Even colorspace conversions can be minimized to 0 depending on the applications and codecs you want to use.

[troyvcd1]

well I have always used CCE to do my dv videos and have had nothing but praise for the quality. I am not an expert and do not have time to modify my scripts. Most people including myself can not tell if it is original or a dvd. I did some tests and now it seems maybe procoder is better though. All CCE videos are min 1000, max 8000, avg 4000 with 3pass vbr Q is at 16 of 64 in CCE 2.67. Procoder is in mastering of course. All are B frames(I dont know if that is the frame type I should be testing or not). I see artifacts in her legs and also in the white "v" above her right shoulder with CCE.
Also C3d did not clean much of this up

Original:


CCE:


CCEwithC3d:


Procoder

[FulciLives]

You want to compare "I" frames when doing this.

- John "FulciLives" Coleman

[BSR]

Well not sure what it is, but CCE and Procoder filtered this part under the girls arm and made it all white.
Where CCE with C3D left it in.


Edit - Oops, shrunk background.
Just noticed the girls thumb. This pics aren't from the same point in time!
Disregard this post!

[troyvcd1]

Why should you compare I frames. Also the avisynth script does delay the video I think half a frame because of the doubleweave.selectodd that is needed for the bff video.

[FulciLives]

Why should you compare I frames. Also the avisynth script does delay the video I think half a frame because of the doubleweave.selectodd that is needed for the bff video.

Because I said so! :P

Adam can explain it better than I if he feels like it.

- John "FulciLives" Coleman

[troyvcd1]

Well if it is the Known way to test mpeg video then I guess I will compare I frames. Also I see that the source video does not match the CCE and procoder video. I used the same frame number from virtualdub but obviously didnt match my videos up. The CCE and procoder clips seem to be at the same place though.

[BJ_M]

some will start at 0 for first frame - some will start at 1

[funkguy4]

some will start at 0 for first frame - some will start at 1

but WHYY!? don't all videos start @ Frame 0???? CCE + AVISynth = awesome sauce

[troyvcd1]

How is it possible to compare I frames when there are so few of them and they happen at different frames for different encoders.

[adam]

Well, for one thing most encoders let you specify exactly where an I frame is placed. You can only place a chapter, for dvd authoring purposes, on an I frame. So manual I-Frame insertion is pretty much a standard feature in mpeg encoders.

And you are going to have at least 1 I frame every second, in most cases, and if you are comparing the same scenes, there is really little change in frames in 1 secs time. So there is no problem in comparing different I frames in the same scene. It is still a much more accurate a comparison then, say comparing the exact same frame # where one encoder makes it an I frame and one makes it a P frame.

You don't necessarily have to compare I frames, but you've got make sure you only compare the same types of frames, (P to P, B to B, or I to I.) They each work in different ways, and P and B frames don't even represent the entire picture. They decode information from different surrounding frames.

[BJ_M]

Well, for one thing most encoders let you specify exactly where an I frame is placed. You can only place a chapter, for dvd authoring purposes, on an I frame. So manual I-Frame insertion is pretty much a standard feature in mpeg encoders.

And you are going to have at least 1 I frame every second, in most cases, and if you are comparing the same scenes, there is really little change in frames in 1 secs time. So there is no problem in comparing different I frames in the same scene. It is still a much more accurate a comparison then, say comparing the exact same frame # where one encoder makes it an I frame and one makes it a P frame.

You don't necessarily have to compare I frames, but you've got make sure you only compare the same types of frames, (P to P, B to B, or I to I.) They each work in different ways, and P and B frames don't even represent the entire picture. They decode information from different surrounding frames.

if they are open GOPs with scene detection on -- the I frames will be on different frames after only a few GOPs (if movement or new scene is detected)

MOST encoders SHOULD have I frame insertion - but many dont ..

tmpgenc 2.x and cce sp (full version) do , vegas IF you use DVDA and insert markers does (not encoding from vegas though , even with markers) ..

Mainconcept and procoder do not - though both do have scene detection ..

[Sulik]

Could someone post graphs of bitrate & quantization from Bitrate Viewer rather than a single frame in a long sequence (you usually can make any encoder look the best depending on which frame you pick) ?

Quantization data is more of a hard fact, assuming there is no pre-processing involved.

[vhelp]

.. those frames that you want to test (and post here for demoing best quality
.. here)
.
Perhaps someone should supply a few brief setups of encoding I-Frames
"specific" to each Encoder brand ??

Here's TMPGenc steps to setup:
(from what I understand, and has ben working for me so far)

* Setting
* GOP structure
* [x] Force picture type setting.. [setting] <-- click this button
* scroll through the timeline to the frame(s) you want I's inserted
.. during encode.
* click OK and proceed to Encode in TMPG.

vdubMOD always reports that exact frame I set to I this way, in it's
time-line info line.

-vhelp 3058

[vhelp]

Mainconcept and procoder do not - though both do have scene detection ..

Even if this is so, would it be feasable to setup these two to encode a given
I-frame by first performing this test:

Assuming the test clip is a 15 second clip, and you can focus TMPG and CCE with
the same I-frame in these two encoder, but in MMC/Procoder, you could

* run a first encode (each encoder)
* then, from this, calculate the frame that leads to an I frame of your prefered
.. test clip pic, if its not the one you want to use throughout the stest w/ all
.. encoder..
* then, advance the starting frame, and re-encode with these two encoders,
.. and repeat until the same I-frame is found, as in TMPG and CCE.

Once the above is complete for both MMC/Procoder, you are now ready to perform
your test encode scenarios of each encoder

-->

There. No reason to not test a given encode.

-vhelp 3059

[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.

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.

Is this from experience? Do you have pics (prefer png) that show these results? The last time I used YMPEG it did well, although these test were done with 3D animation from Hash A:M. Procoder is good but blurs, CCE was too noisy and TMPGenc did well.

I guess I'll do some more test some time but with DV.

[BJ_M]

Mainconcept and procoder do not - though both do have scene detection ..

Even if this is so, would it be feasable to setup these two to encode a given
I-frame by first performing this test:

Assuming the test clip is a 15 second clip, and you can focus TMPG and CCE with
the same I-frame in these two encoder, but in MMC/Procoder, you could

* run a first encode (each encoder)
* then, from this, calculate the frame that leads to an I frame of your prefered
.. test clip pic, if its not the one you want to use throughout the stest w/ all
.. encoder..
* then, advance the starting frame, and re-encode with these two encoders,
.. and repeat until the same I-frame is found, as in TMPG and CCE.

Once the above is complete for both MMC/Procoder, you are now ready to perform
your test encode scenarios of each encoder

-->

There. No reason to not test a given encode.

-vhelp 3059

yes -- it could be done that way

[vhelp]

Assuming that we are talking from sources from "captures", I agree w/ it being
from DV. One of the main reasons why, is because:

A) It is the most cleanest source available. Analog cards exhibit too much
.. Noise into the captured source. This can be from any number of phenominas.

B) It is the least source to have issues (ie, Frame Drops; Audio sync, etc)

C) And, because its here to stay for a long time, and so far, I have no
.. plans of leaving it. Not w/ out my ADVC-100 you know. hehe..

@ BJ_M

I'm glad we seem to agree on something here
.
I think that you should not have any more lip-back from anyone in your next
upload of tests

Cheers,
-vhelp 3060

[FredThompson]

Just how do you propose to get an analog signal to a DV encoder? analog sampling in DV isn't inherently cleaner than that of an analog capture card. Design, construction, components and environment all play a part.

wrt DV being wonderful, here are some unfiltered excerpts of DV camcorder source. The first is PAL (Sony DCR-HC14E), the second is NTSC (Canon ZR85MC). Saturation is a little light in the NTSC but way overdone in the PAL. Which does a better job with diagonal lines?

[lordsmurf]

CCE adds noise.
If you want to see "soft" use some of those free MPEG encoders, or MainConcept standalone version.

This is why I use Procoder a lot.
TMPGENC if I need quality filters to restore.

[waltzingandy]

Hi,

I'm fairly new to all this in that I've been lucky enough to use a couple of editiing tools (Pinnacle 8 and then Premiere Pro) early on and now I've got hold of Pinnacle Liquid Edition. I was just wondering if anybody knows if the LE encoder engine is Pinnacles or did they license it, and how it stacks up in these discussions???

I did do my own (very basic, just simple viewing of the footage...) testing some time ago and it seamed that Pinnacles output was as good as anyone elses? Comments?

If noone has access to LE I guess I could get the source DV and enocde and send the results to somebody so they can do comparisons (I must admit, the I frame discussions are a little beyond me )!

Thanks,
Andy.

[Rough]

I agree with Baker. I also use YMPEG and interested to know comparison results with YMPEG, if there are any. I encoded fire avi using YMPEG 2.5 with surprisingly superb results (1500000, 6000000, 8000000 bps, progressive settings) , much better than CCE etc.

May I suggest to include this encoder in test without any prejudice. This nifty but not much noticed encoder worth giving a try.

[BJ_M]

I agree with Baker. I also use YMPEG and interested to know comparison results with YMPEG, if there are any. I encoded fire avi using YMPEG 2.5 with surprisingly superb results (1500000, 6000000, 8000000 bps, progressive settings) , much better than CCE etc.

May I suggest to include this encoder in test without any prejudice. This nifty but not much noticed encoder worth giving a try.

i think speed was its only drawback really -- there is a new version out now .though ... it is $25

[Rough]

watch out for YMPEG bitrate settings if it is too slow. Oddly enough, YMPEG uses bps for all bitrate values (and not kbps). This is sometime misleading. If you enter say 6000 for bitrate value thinking that it is in kbps, YMPEG will silently accept it and starts encoding at 6kbps (instead if 6000 kbps as one intended) and in that case it will be very slow. I do not see speed as a problem otherwise. As a matter of fact, I find it faster than TMPGEnc

Checkout a thread in KVCD forum discussion this.

[guns1inger]

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

A couple of friends and I fancy ourselves as amateur wine buffs. We know the local product fairly well, and have sampled (often in very comfortable quantities) most wines our country has to offer (sorry, Napa Valley, but you can keep your watery cat's piss), from $12 quaffing plonk, up to cellered hermatige in the $100's. We generally settle on the $25-35 mark most of the time though. We acknowledge that the hermatige is better than the $30 bottles I have in my celler, but it is not the 10 - 12 times better than the price would indicate.

I see encoders in a similar way. I have read LordSmurf's arguments against CCE and the noise it adds. I have read this thread, and many like it. I have used tmpgenc 2.5, procoder 2.0, mainconcept (both standalone and vegas) and CCE. Yes, both tmpgenc and procoder have nice filters, and yes, both can produce nice output using these. But a job I can complete in 3 hours in CCE takes 19 in procoder and 27 in tmpgenc, on the same machine, and using the same settings. The only difference is the engines, and the filters. Take the filters off and times come down (to 12 and 18 hours respectively), 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.