digitalFAQ.com Forums [Archives] - TMPGEnc: Sample of new TMPGEnc 2.524.63.181

digitalFAQ.com Forums [Archives] (http://www.digitalfaq.com/archives/)

- Video Encoding and Conversion (http://www.digitalfaq.com/archives/encode/)

- - TMPGEnc: Sample of new TMPGEnc 2.524.63.181 (http://www.digitalfaq.com/archives/encode/13053-tmpgenc-sample-tmpgenc.html)

kwag	12-18-2004 12:30 AM

Sample of new TMPGEnc 2.524.63.181

Here's an MPEG-1 sample of what the new encoder does:

http://www.kvcd.net/downloads/NewScript.m1v.cut.40.mpg

MPEG-1 704x480 23.976fps CQ 70.5913848876953 and GOP = 24

Script used:

Code:

Mpeg2Source("H:\RED_PLANET\VIDEO_TS\red.d2v")

Undot()

Limiter()



BilinearResize(672, 254, 3, 59, 714, 360)





## Linear Motion Adaptive Filtering ##

ScriptClip("nf = YDifferenceToNext()" + chr(13) + " \

UnFilter( -(fmin(round(nf)*2, 100)), -(fmin(round(nf)*2, 100)) ). \

TemporalSoften( fmin(round(2/nf),6), round(1/nf), round(3/nf), 1, 1) ")



AddBorders(16, 113, 16, 113)

Limiter()



## Functions ##

function fmin(int a, int b) {return (a < b) ? a : b}

Note: The complete video, calculated with CQMatic, gave a final size of 758,800KB. Wanted target was 719,486KB.
The last four minutes of this movie are end credits :!:
This is probably the cause of the slight overshoot of the video size.
I am now going to do some investigations on CQMatic to take into consideration start and end credits ;)

-kwag

muaddib

12-18-2004 01:13 AM

Wow, kwag!
What a great sample! :D
It's amazing how a movie like this can fit in a single CDR with THAT QUALITY!!! 8)

Shibblet

12-18-2004 02:21 AM

Borders

Hey Kwaggy,

Quote:

AddBorders(16, 113, 16, 113)

If you make your borders all divisible by 8, you will compress better. Try it with top and bottom borders of 112.

One pixel won't cause that much of an issue, but TMPEG compresses black better if the block is in pixel dimensions of 8x8.
[/quote]

Also, is there any reason why you use Red Planet for your tests? Does it have a lot of motion, or does it compress well?

kwag	12-18-2004 02:37 AM

Re: Borders

Quote:

Originally Posted by Shibblet

If you make your borders all divisible by 8, you will compress better. Try it with top and bottom borders of 112.

Look at the aspect ratio percent error in Moviestacker or FitCD, when you divide by 2 and by 8 ;)

Quote:

One pixel won't cause that much of an issue, but TMPEG compresses black better if the block is in pixel dimensions of 8x8.

True. But I care more about the correct aspect ratio. The final difference in size is not really that much from 2 to 8. Have you tried it :?:
Just encode some 5 minute clip or so, with a precision of 2, and then at 8. See what difference you get in file size.

Quote:

Also, is there any reason why you use Red Planet for your tests? Does it have a lot of motion, or does it compress well?

I've been using Red Planet as a reference for a very long time :)
It's just a way to maintain consistency, so I can see the different qualities created by different encoders.
It's not because of some special aspect of the movie.

-kwag

Shibblet

12-18-2004 03:11 AM

Quote:

Look at the aspect ratio percent error in Moviestacker or FitCD, when you divide by 2 and by 8

Yeah, that's pretty messed up.

Quote:

True. But I care more about the correct aspect ratio. The final difference in size is not really that much from 2 to 8. Have you tried it
Just encode some 5 minute clip or so, with a precision of 2, and then at 8. See what difference you get in file size.

Yeah, I do it with all of my movies, and you only have to compensate for a maxiumum of 4 pixels either way. Secondly, if you decide to crop just a little bit more (like a pixel or two) off of the ends, you can adjust right to the 8 (1 block) mark. :P

It also seems to get rid of that weird blurry top line on most widescreen movies. 8)

Code:

BilinearResize(672, 256, 3, 59, 714, 360)

AddBorders(16, 112, 16, 112)

Also, any reason you use Bilinear resizing as opposed to Bicubic? Or is that another one of those 6 of one, half-a-dozen of another?

kwag	12-18-2004 03:17 AM

Quote:

Originally Posted by Shibblet

Also, any reason you use Bilinear resizing as opposed to Bicubic? Or is that another one of those 6 of one, half-a-dozen of another?

I screwed up on that one :!:
I always use Bicubic over bilinear EVERY time :!:
Thanks for pointing that out.
No wonder my video was not as sharp :oops:
Time to re-run CQMatic and do it again :x

Thanks!,
-kwag

incredible

12-18-2004 04:54 AM

Quote:

If you make your borders all divisible by 8, you will compress better. Try it with top and bottom borders of 112.

As we do encode in mpeg which means ...

4:2:2 = YV12 = Half vertical and half horizontal chroma resolution, we need a total Image architecture of MOD16 so the chroma will end up in MOD8 (16/halfChromaFreq).

When setting up FitCD/Moviestacker use when cropping progressive Stuf in YV12 "round by 2" and in case of interlaced "round by 4"

But all that only results ion its purpose IF the image is placed "Macroblock orientated" afterwards to match the MOD16 Matrix.
But macroblock orient. placement is not everybodys "gusto" ;)

Dialhot

12-18-2004 10:03 AM

Karl I'm waiting for your sample with bicubic. I always use Lanczos personally and bilinear is a too big step to cross for my eyes :-). I can't appreciate fully your sample because of this :-D

kwag	12-18-2004 10:07 AM

Quote:

Originally Posted by Dialhot

Karl I'm waiting for your sample with bicubic.

I woke up to a crashed machine this morning :x
I've had it with XP(ig) :!:
I'm installing 2000 Professional this week, but first I have to make a BIG backup :!:
Anyway, I've started encoding again. I'll put up the new sample as soon as it's done.

Quote:

I always use Lanczos personally and bilinear is a too big step to cross for my eyes :-). I can't appreciate fully your sample because of this :-D

I know :roll:

-kwag

Dialhot

12-18-2004 10:11 AM

Note : can you tell us the CQ use ?

I don't want to screw up your dreams but there are a lot of defaults in this sample...

(little but noticiable mosquitos on faces and arround the name stamped on the suits at second 11 for instance...)

kwag	12-18-2004 10:36 AM

Quote:

Originally Posted by Dialhot

Note : can you tell us the CQ use ?

It's in the post :)
CQ 70.5913848876953

And remember, this was just a test WITHOUT any big filtering. Just MA section. This DVD is not 100% clean. There are small flaws on the VOBs, so that's why the encoding is not perfect.

Quote:

I don't want to screw up your dreams but there are a lot of defaults in this sample...

(little but noticiable mosquitos on faces and arround the name stamped on the suits at second 11 for instance...)

True, but I'd like to see if CCE or other encoders can clean those mosquitos at that low bitrate, specially because the movie is 101 minutes long, and the target I calculated is for one CD ;)

Edit: Encoder started at CQ=70.2341766357422
This seems perfect, as the end credits were cut of prior to prediction, so this should now match wanted file size almost perfect.
Here's a sample of the "sampler", because full encode just started now.
http://www.kvcd.net/downloads/NewScr....m1v.cut.1.mpg

-kwag

kwag	12-18-2004 11:49 AM

Opinions :?:
Good or bad, are always welcome ;)

-kwag

Dialhot

12-18-2004 12:28 PM

Quote:

Originally Posted by kwag

And remember this was just a test WITHOUT any big filtering. Just MA section. This DVD is not 100% clean. There are small flaws on the VOBs, so that's why the encoding is not perfect.

I saw, and that's also why I asked for the CQ. 70 is not bad but near the limit for an "acceptable" KVCD. That explains the visual defaults.

Quote:

True, but I'd like to see if CCE or other encoders can clean those mosquitos at that low bitrate, specially because the movie is 101 minutes long, and the target I calculated is for one CD ;)

I don't think they can. The defaults are due to too light filtering.

EDIT: tha sample is more pleasant to my eyes but of course, sharper resizer means much more mosquitoes :-( (why don't you do your test with the full MA script ?).
What is awesome is the lack of DCT blocks in plain areas.

kwag	12-18-2004 01:00 PM

Quote:

Originally Posted by Dialhot

(why don't you do your test with the full MA script ?).

Shure will :)
But today i'm working on CQMatic ;)

Quote:

What is awesome is the lack of DCT blocks in plain areas.

It sure is :)
I definitely see some improvements on TMPEG :cool:

-kwag

muaddib

12-19-2004 01:22 AM

Quote:

Originally Posted by kwag

Opinions :?:
Good or bad, are always welcome ;)

I still think that the first one is a great sample... as it would fit a 704x480 in a single CD with that quality!! :D
About the second sample... well, I have mixed feelings.
I agree that the first one was a bit too soft, and the sharper image sure is more pleasant, but there are much more "mosquitoes" in this last sample.
I'm not sure.... I would like to see a sample with the same cut as the first one, but I think that I prefer the soft image to the mosquitoes.

Quote:

Originally Posted by Dialhot

What is awesome is the lack of DCT blocks in plain areas.

Exact! That's awesome! Especially because, in my opinion, DCT blocks are one of the ugliest problems of a high compressed encode.

jeo	12-20-2004 05:27 AM

Quote:

Originally Posted by kwag

Opinions :?:
Good or bad, are always welcome ;)

-kwag

Good.....really really good....
(i saw the first sample ony, downloading the second)
8)
my next target is Kdvds mpeg1! that idea is growing HERE

jeo	12-20-2004 05:43 AM

ideas(seems good) growing! :idea:

CQ90 for 3 (or 4) movies mpeg1 720*xxx for Kdvds, encode the audio in the end after see the free space to choose the bitrate and forget perfect calculations about final size for KDVDs! (headac3he will show the final size before encode each track), is clear?

:arrow: what you think boys :?:

Dialhot

12-20-2004 09:49 AM

Quote:

Originally Posted by jeo

:arrow: what you think boys :?:

You should see on the other thread why I said about beeing carefull with MPEG1 KDVD : no pulldown and no anamorphic disc...
(please, continue the discussion in the other thread).

rendalunit

12-28-2004 11:58 PM

Re: Sample of new TMPGEnc 2.524.63.181

Quote:

Originally Posted by kwag

CQ 70.5913848876953

Well it's precise enough! How long did it take to come up with that value :?:

kwag	12-29-2004 10:11 AM

Re: Sample of new TMPGEnc 2.524.63.181

Quote:

Originally Posted by rendalunit

Quote:

Originally Posted by kwag

CQ 70.5913848876953

Well it's precise enough! How long did it take to come up with that value :?:

Around 25 to 30 minutes.

-kwag