Correct Aspect-Ratio in x264?

[amango]

I used XVID all the time to encode PAL DVB-streams - now I want to test x264.

Most streams I encode are in 720x576. In XVID I selected the 4:3-PAL setting for the Aspect-Ratio, so the video will be stretched correctly during playback.

In x264 I have the option "Sample AR".

I would like to know what settings I need to type in for correct 4:3-PAL and 16:9-PAL in x264.

[stephanV]

4:3 PAL = 12:11
16:9 PAL = 16:11

If you use these settings you are doing it the same way in x264 and XviD.

[amango]

Thank you! :)

[SeeMoreDigital]

Or......

PAL 4:3 DAR = 16:15
PAL 16:9 DAR = 64:45

NTSC 4:3 DAR = 8:9
NTSC 16:9 DAR = 32:27

EDIT: In February 2006 I conceived a tool called ARS Calculator (http://forum.doom9.org/showthread.php?t=107039) to help people work out Aspect Ratio Signalling (ARS) ratios (ie: PAR, SAR, DAR) for use with encoders such x264 and XviD


Cheers

[amango]

What's the formula...?

Sometimes I have DVB-streams in 480x576 or 704x576 PAL. Can I also set the correct AR in x264 for this resolutions...?

[SeeMoreDigital]

Personally speaking I prefer to use the (4:3 or) 16:9 DAR calculations because they will always output an (4:3 or) 16:9 shaped image, regardless of the sources pixel size/ratio.


Cheers

[stephanV]

What's the formula...?

Sometimes I have DVB-streams in 480x576 or 704x576 PAL. Can I also set the correct AR in x264 for this resolutions...?
PAR (or SAR) = DAR*height/width

[fatxy]

which player besides vlc can handle that properly?

tested zoomplayer with ffdshow without success

[stephanV]

enable overlay mixer in ffdshow

[fatxy]

thanks i changed the output to overlay mixer for both ffdshow and zplayer
i only get a black screen :/

[Sharktooth]

Added more info on SAR in the daily builds FAQs.

[stephanV]

thanks i changed the output to overlay mixer for both ffdshow and zplayer
i only get a black screen :/
Thats sucks... have you tried MPC?

[JoeBG]

thanks i changed the output to overlay mixer for both ffdshow and zplayer
i only get a black screen :/

overlay mixer and ffdshow work perfect in zplayer. You can also take Video Mixing Renderer 7 and ffdshow. Looks great :)

[SeeMoreDigital]

Just in-case anybody needs to know, here are the FFdshow "overlay settings" I use in-conjunction with MediaPlayer Classic: -

http://img108.imageshack.us/img108/66/ffdshowanamorphicoverlaysettin.png

Encodes with AR signalling in AVI, MP4, MKV and OGM work perfectly....


Cheers

[fatxy]

ok thanks a lot

zplayer vmr and ffdshow overlay mixer works (thought i have to use it for both), not as smooth as vcl but its workin (i have to click on fit source one time)

besides i dont really know what to do with those crappy 480x576 streams

480x576 autozoom to 768x576

or somethin like

480x576 encode to 640x480 or even 768x576

[Inc]

PAR (or SAR) = DAR*height/width

In case of respecting the actual "real" DAR of your source then it makes sense.

Good Example a 52.000µs capture:
DAR = 1,3333 * 576 / 702 = almost 1,094:1 = PAR 128:117
Here a 702x576 is truely DAR 4:3 @ PAR 1,094

Bad example, a fullscreen 720x576 DVD as source:
DAR = 1,3333 * 576/720 = 1,06664 = the common error of generic par happens here

720 * (128/117) / 576 = DAR 1,3675:1 equals not 4:3 (1,3333)

DAR = 1,3675 * 576 / 720 = PAR 1,094:1 = ok

:)



480x576 autozoom to 768x576

or somethin like

480x576 encode to 640x480 or even 768x576

:eek:

480x576 px sources do have 53,3333us (same as 720x576).
768x576 do have 52,0000us

53,3333/52,0000 = 1,02564 <--- our diff. factor!

So do resize 480 x 576 to 768*(1,02564) x 576 = 788x576

or do crop 480 x 576 to 480/1,02564 x 576 = 468x576 and resize to 768x576

;)

If you do respect these your eyes when watching mpeg4s via HTPC will kiss you.

[SeeMoreDigital]

Lookie here... I've made some Anamorphic Test Files (http://81.98.148.105/Uploaded_Files/Doom9_Forum_files/x264_Anamorphic_Test_Files.7z) for you guys to try!

Provided you have the correct decoder filters installed they should all open up at 16:9...


Cheers

[Piper]

@Incredible:

Good Example a 52.000µs capture:
DAR = 1,3333 * 576 / 702 = almost 1,094:1 = PAR 128:117
Here a 702x576 is truely DAR 4:3 @ PAR 1,094

The only part I don't follow is how you derive PAR 128:117 from 1,094:1. What's the calculation being used here?

Thanks

[Inc]

Full TV PAL is 768x576 @ 52.000µs @ 14,7692307Mhz (= quadra pixels)
ITU samplingrate is 13,5 Mhz (= squarepixels)

14,7692307/13,5 = 1,094017 = TV PAR @ 13.5Mhz used by a SAP

or ...
52.000µs * 13.5Mhz = 702 active pixels at square pixels
52.000µs * 14,7692307Mhz = 768 active pixels at quadra pixels

768/702 equals 128/117 equals 1,094017

;)

[SeeMoreDigital]

That's fine for anamorphic encodes with pixel frame width of 702....

However, if you want to generate spec PAL video files with a target aspect ratio of 4:3 based on a pixel frame width of 720, I would recommend you use the following calculation: -

768/720 = Lowest fraction = 16/15 (Decimal = 1.094017)


However, if you want to generate spec PAL video files with a target aspect ratio of 16:9 based on a pixel frame width of 720, I would recommend you use the following calculation: -

1024/720 = Lowest fraction = 64/45 (Decimal = 1.422222)


Cheers

[Piper]

What would you suggest for NTSC Analogue captures with a width of 720 or 704 pixels?

For example, I have an analogue capture at 720x480 which I crop to 720x360 to remove letterboxing, the resize back to 720x480 anamorphic.

Using the cropped resolution:
16.9 = 1.7777 -> (1.7777x360)/720 = ~640/720 = 0.888

640/720 equals 8/9 equals 0.8888

This doesn't look right.

EDIT: This of course assumes the original pixel width is actually 720 and not at all taking into account things like what my ATI AIW 8500's capture window is etc etc.

[Inc]

That's fine for anamorphic encodes with pixel frame width of 702....

However, if you want to generate spec PAL video files with a target aspect ratio of 4:3 based on a pixel frame width of 720, I would recommend you use the following calculation: -

768/720 = Lowest fraction = 16/15 (Decimal = 1.094017)
16/15 = 1,066666667 our known "generic PAR" trap
768 are within 702, so 768/702 = Lowest fraction = 128/117 = and THAT results in 1.094017

However, if you want to generate spec PAL video files with a target aspect ratio of 16:9 based on a pixel frame width of 720, I would recommend you use the following calculation: -

1024/720 = Lowest fraction = 64/45 (Decimal = 1.422222)

720*(128/117)*1.3333 = 1050.23! <--- = PAR 1:1 non anamorph state out of an anamorph PAL 720 stream.
If you would like to end up in 1024 width, then you have to crop from 720 to 702 before:
1050,23/1024 = 1,02561523 (=pre cropping factor)
720/1,02561523 = 702,018 = ~702 ---> BINGO

So the anamorph PAR is 1024/702 = 1,458689

Have a look at "PARanoia" as it does respect those specs in its calcs.

Your examples above do assume that a source window of 52.000µs was brought into 720, as that results sometimes out of wrong capturing settings, but as *we do hope* that almost all industry 720 sources (like DVDs) do result out of a 53.333µs sampling, then 768 are within 702.

You also can compensate in PARanoia such 720@generic PAR source-errors.

640/720 equals 8/9 equals 0.8888

??

640 in NTSC equals 702 where FULL NTSC is 648 which equals ~711
So DONT scale 720 to 640! (assuming your 720's source has been ITU conform mastered)

[SeeMoreDigital]

640/720 equals 8/9 equals 0.8888

This doesn't look right.I think you'll find it does though (http://81.98.148.105/Uploaded_Files/Doom9_Forum_files/x264@720x360_anamorphic_samples.7z) ;)


Cheers

[lexor]

why do you guys care about apect ration settings? I never used them I never had playback issues, output is same aspect ratio and resolution as input (input being avisynth resize)

[Piper]

When encoding a source as anamorphic, then knowing the AR is a requirement. If simply cropping (letterbox), then sure - leave it as is.

[Piper]

I think you'll find it does though ;)

Say for arguments sake I captured my original source at 640x480. Given the above logic this would translate to (when cropped to 360 vertical):

1.7777x360/640 = ~640/640 = 1

In this case, I would presume I wouldn't have to make any changes to SAR (if encoding to x264 and SAR 1:1 is default). If it were resized then from 640x360 to 640x480 anamorphic, I would explicitly have to set the AR to 16:9 when playing it back no?

[SeeMoreDigital]

Say for arguments sake I captured my original source at 640x480. To start with... Would this source have originally been encoded with "square" pixels or "anamorphic" pixels?

What would an uncorrected "still" image taken from this source look like?

[Piper]

I'll have to dig around to see if I still have an unmodified source to produce a still from. Typical sources for which I'd apply such treatment to are: Doctor Who and Coronation Street <--don't laugh. :D These are originally PAL sources which are rebroadcast on CBC (NTSC). Whether they're priginally square pixel or anamorphic I don't know. I do know these series are also broadcast on CBC HD.

Back to my earlier statment:

640/720 equals 8/9 equals 0.8888

This doesn't look right.

A SAR of 8:9 is 4:3!

[Inc]

What would you suggest for NTSC Analogue captures with a width of 720 or 704 pixels?

For example, I have an analogue capture at 720x480 which I crop to 720x360 to remove letterboxing, the resize back to 720x480 anamorphic.
Say for arguments sake I captured my original source at 640x480. Given the above logic this would translate to (when cropped to 360 vertical):

1. Whats your capture card, means model/brand, then we do see if really the 720 do have a 53.333µs window. Means IF your 720px do have thin black lines on the right and left, then they seem to be captured correct.

2. Captured out of an analogue broadcasting signal arent anamorph but sometimes letterboxed if the source was 16:9.

3. There is only ONE of your capture resulotions correct, but both cant be.
Means 640x480 do have a par of 1:1 at 52.000µs, 720x480 do have regulary 53.333µs, so one of your both capture sizes did result wrong imho as I dont know a driver which is smart related to individual capping resolutions.

4. If your capture source is 640x480 (which is par 1:1) and you do crop the borders so it becomes 640x320, then if encoding to mp4/xvid/Divx etc. do encode in THAT resolution as the par is already 1:1 and everything btw. is at Mod16.

640*(1/1)/320 = 2, your DAR then is 2:1

If you want to bring that to an anamorph state, then:

640 x 320*(1,3333) = 640 x 427

[SeeMoreDigital]

Hi Piper,

I think you need to do a little bit more reading up about when and why "anamorphic" images are used.... and look they like before correction!


Cheers

[Piper]

1. Whats your capture card, means model/brand, then we do see if really the 720 do have a 53.333µs window. Means IF your 720px do have thin black lines on the right and left, then they seem to be captured correct.

It's an ATI All-In-Wonder 8500 128Mb

Yes, when I capture at 720x480 there are 8 black pixels on either side. I only suggested 640x480 for arguments sake, I never do actually capture at that resolution. I do however often crop 720 to 704 however. In this case my cropped resolution would be 704x360 which would give...

640/704 equals 10/11 equals 0.9090

My destination format is typically DVD, hence the 720 or 704/480 anamorphic resolution. EDIT: For DVD of course, setting the 16:9 flag is all that's required. It's what settings to use when encoding x264 that I'm attempting to clarify.

[Piper]

Hi Piper,

I think you need to do a little bit more reading up about when and why "anamorphic" images are used.... and look they like before correction!


Cheers


Agreed. Thank you both for your input!

[Piper]

I think I'm getting somewhere with this.

After some searching around, I re-discovered this page: http://www.doom9.org/index.html?/aspectratios.htm

If I apply the 16:9 calculation it to my 704x360 cropped source we end up with:

360*16/9 = 640
640/480 (maximum vertical resolution) = 1.3333 = 4/3 SAR

Comparing the output from my x264 encode (--sar 4:3) to the original unmodified clip in VirtualDub, the AR is indeed identical to each other. :D

*After re-reading the aspectratios page I realised I had made the assumption to divide 640/480, rather than by its width as I had read in other calculations.

EDIT: Correction

[Inc]

Crop your 720x480 to 702x480(360) and resize to 640x480(360).

Or as said, do use PARanoia as it does the calcs for you, it also supports re-anamorphing a letterboxed source.

[SeeMoreDigital]

Are you able to provide an corrected still image of your source?

For all we know the "image" pixels might run the full width edge-to-edge!


Cheers

[Inc]

For all we know the "image" pixels might run the full width edge-to-edge!
Seems his chip digitizes at 53.333µs so the window in "his" 720px seem to be ITU compilant:
Yes, when I capture at 720x480 there are 8 black pixels on either side.

But providing image samples are never that bad :)

@Piper
Do use http://www.imageshack.us for providing your sample online.
(Don't use the resize option at imageshack for shure)

[Piper]

Ok, here's a sample frame.

http://img16.imageshack.us/img16/6883/capsample1la.th.jpg (http://img16.imageshack.us/my.php?image=capsample1la.jpg)

I've run it through PARanoia (nice app Incredible!) and I ran Autocrop samples, chose 704x480 resizing (I'd like to stay with dvd compliant resolutions), and chose Nonanamorph to Anamorph. I also tweaked the Y-Pos to 60 and Height to 360. Autocrop samples otherwise included the bottom half of the olympic symbol in the CBC logo.

It also looks like my 8 black pixels on either side was off. Autocrop chose 5 pixels on the left, 3 on the right.

With this, target PAR becomes 1.21557:1. Assuming this is correct, what's the formula for changing 1.21557:1 into something I can enter into SAR in MeGui? I'm sure I already know this....it's been a long day....

Thanks

[Inc]

I did a small AVI from your provided pic.

When opening PARAnoia for the first time you first have to set the active capturearea to your cards specs.
As I do assume your card samples using 53.3333us so we have to set that up:

http://tinypic.com/magids.gif



Then as you will end up in a resizing for mpeg4 encoding purposes, you have to set the mpeg4 size profile:
(I used standard 720x576!! for this workout!)

http://tinypic.com/magigl.gif



With a little tweaking of the autocropping I receive this result:

http://tinypic.com/magimu.jpg

You can see the target PAR will be 1.3333



Lets preview the script:

http://tinypic.com/magioh.jpg



And lets check by resizing that preview in Photoshop using the PAR of 1.33333:

http://tinypic.com/magiv8.jpg

Voilá

Do a little test encode in MeGUI using in the resulted script the Trim() commands. Then set in the x264 config dialogue the PAR of 1.3333 as SAR. Then we do see if in x264 the encode using this (PAR as) SAR results in a later correct playback.

PS: Btw. you can see that "re-anamorphing" this letterboxed capture source doesn't really make sense as the imagebroadcasting already comes in a quite bad condition ;)

[bkman]

The SAR for 704 x 480 (4:3 DAR) is 10x11 ;)

(I think someone was asking this)

[SeeMoreDigital]

When you "hard crop" away the mattes from the 720x480 still image you provided, you are left with an 712x362 pixel image frame size. Therefore the nearest "mod 16" calculation for both axis would be 720x368.

Over here in the UK, the new series of Dr Who is predominately shot on Digi-beta and release in true 16:9 (aka 1.7777) wide-srceen TV format.

So... if you use the "mod 16" height of 368 pixels as a start point and multiply it by 1.7777 you obtain an "anamorphic" target width of 654.2222 pixels.

If you then divided 654.2222 by 720 you get 0.9086 (as a decimal), which equates to 189:208 as a fraction or PAR ratio. The difficulty is converting decimals into fractions.

Or if you're okay at working with fractions, try this approach. Which is more accurate: -

16 368 5888 368
-- x --- = ---- lowest dominator --- giving a PAR of 368:405
9 720 6480 405


Anyway, here's a 10 second sample (http://81.98.148.105/Uploaded_Files/Doom9_Forum_files/SMD_AVC_AR_Sample_01.zip).


Cheers

[Inc]

@bkman

the correct PAR of a 720x480 NTSC 525 Lines video stream is 4320/4739, which results in 0,9115847.
Your resulted SAR is 0,909091 out of 10:11

ITU 720x480 uses 13.5mhz
A PC Monitor equals to 12 + 1452/4739² Mhz

So the result is:
( 12 + 1452/4739² ) / 13.5 = 0,9115847 = 4320/4739 ... Bingo

These penible calcs do prevent a NTSC resizing to 640x480 beeing 2px off finally *lol*
But these numbers above only try to explain the logic.
When using generic PAR then you will be in much more recognisable trouble, means if on 720x576 a PAR of
1,0666667 is used.

[Inc]

... and released in true 16:9 (aka 1.7777) wide-srceen TV format.
True 16:9 in a analogue broadcasting? Even if thats possible, then why does his pic sample comes in nonanamorph?
712(*NTSC 720_PAR) / 362 = 1,79295 and not 1.77777

No offense, but I looked at your sample and the circle below in the station Logo isnt "round" (VLC latest release was used for playback, so Im not shure if it displays correct, theplayback of "your" circle is a tiny bit unround).

This it "how" it should look if a correct PAR was/is used:

http://img30.imageshack.us/img30/8286/f3ie.gif (http://imageshack.us)

[bkman]

@bkman

the correct PAR of a 720x480 NTSC 525 Lines video stream is 4320/4739, which results in 0,9115847.
Your resulted SAR is 0,909091 out of 10:11

I was giving the PAR for 704x480, not 720. Ie, 8 (black) pixels on either side cropped away. This SAR results in a 4:3 DAR, which is what we want if the source was 4:3, no? Or am I being far too naive about this..?

[Inc]

Everything depends on HOW the source is digitized.
Means IF "he" uses a card which samples 720x480 using 53.333us, then the PAR is ITU compilant NTSC D1 of 4320/4739 (like on DVDs). If you crop to 704x480 then the DAR will be changed but NOT the PAR as the PixelAspectRatio will be the same. 720x480 at ITU isnt 4:3 neither 704 (if beeing correct) its 702 out of 720.

[SeeMoreDigital]

So this does not look okay to you then!

http://img520.imageshack.us/img520/5452/smdround1it.png


Cheers

[Inc]

Yes in here it does ;)

[Piper]

@SMD: If you then divided 654.2222 by 720 you get 0.9086 (as a decimal), which equates to 189:208 as a fraction or PAR ratio. The difficulty is converting decimals into fractions

I found a handy site for converting decimals: Fast Decimal to Fraction Approximations (http://www.mindspring.com/%7Ealanh/fracs.html)

I did my own calculation using the same 712x362 in PARanoia and came up with a DAR of 1.783 => PAR 0.9065 => 223:246. The first time I ran this I applied SAR 223 x 246 in MeGUI which stretched the vertical. Reversing the two numbers in MeGUI to read 246 x 223 produced something very similar to Incredible's 0.9115847 PAR = 229:328 and SMD's 189:208 (if likewise reversed in MeGUI). So who's circle is the roundest then? heh

Am I reading this incorrectly? If PAR (SAR) is say PAR 0.9065 => 223:246 then wouldn't it make sense to input it in MeGUI as SAR 223 x 246? Or better still, perhaps Sharktooth could modify it to allow a decimal number.

@SMD: Your sample clip is an odd resolution isn't it? Why resize?

EDIT: Entire post

[SeeMoreDigital]

@SMD: Your sample clip is an odd resolution isn't it? Why resize?As I mentioned before, the pixel size of my video sample was derived from the image you provided and was "resized" to the nearest 16th pixel, ie: "mod 16".

A pixel size of 712x362 does not equate to "mod16" (712 equates to "mod8" and 362 equates to "mod2"). So calculating a PAR from it would not be very wise.

And many thanks for providing the "Decimals to Fractions" link.... I will check that out ;)


Cheers

[Inc]

Yep, thats a nice Hint, so Ill apply to PARanoia a box or stringfield where the correct fraction for X264 can be seen.

[Piper]

A pixel size of 712x362 does not equate to "mod16" (712 equates to "mod8" and 362 equates to "mod2"). So calculating a PAR from it would not be very wise.

Ah yes of course. Thanks for pointing that out (again).

Thanks again SMD & Incredible for your assistance in this thread, it's very much appreciated. :D

[lenouch]

Is PARanoia a freeware ? If yes, anyone got a link to download it? It looks nice, but google always give me results concering the "true" word paranoia

thanks :)

[SeeMoreDigital]

If you had used the forums search facility you would have found this: -

http://forum.doom9.org/showthread.php?t=101070&highlight=PARanoia

[Rash]

Look, I have just one question that may seem very stupid after all this great discussion. :) I've read it all but I couldn't extract my question out of it.

Let's see if I can make this simple. I have a 720x480 video from a DVD. The video is flagged as anamorphic so the real video is DAR 16:9. The DVD player would then display the video as 853x480 (forget about mod16 just for now), correct so far?

So this video has black bars, and I crop it. The cropped resolution is 720x352. My question is: which DAR do I use to calculate SAR for this cropped resolution? 4:3 from the source video (720x480), 16:9 that is the real DAR for the source video (853x480) or the cropped resolution DAR 2.05:1 (720x352)?

Thank you a lot for your attention. ;)

[SeeMoreDigital]

...So this video has black bars, and I crop it. The cropped resolution is 720x352. My question is: which DAR do I use to calculate SAR for this cropped resolution? 4:3 from the source video (720x480), 16:9 that is the real DAR for the source video (853x480) or the cropped resolution DAR 2.05:1 (720x352)?Try: -

68:75 - If the "Movies" AR is 1.85:1

54:47 - If the "Movies" AR is 2.35:1

88:75 - If the "Movies" AR is 2.40:1

EDIT: By-the-way.... what is the movies" aspect ratio. It should tell you somewhere on the DVD case. If you have not got the case you should be able to check it over on Amazon.com


Cheers

[Inc]

The video is flagged as anamorphic so the real video is DAR 16:9. No thats just the flag in the streams header so the SAP knows if letterboxing is needed if a 4:3 TV set is the target device. ;)
The DVD player would then display the video as 853x480 (forget about mod16 just for now), correct so far?
The DVD Player just sends a Signal @13.5Mhz wich equals 720x480px
On a "16:9" Tv it would be displayed as this:

NTSC_D1_PAR = 4320/4739

So 720*(4320/4739) x 480 = 656,341 x 480 (still anamorph, means squeezed)
Lets do take the anamorph factor into account:
656,341*(1,33333) x 480 = 875 x 480 ... so not 853x480 ;)
So your 720x480's real DAR is 875/480 = 1,823
So your cropped stream results in 875 x 352 = a real DAR of 2,486 ... use this one to calculate the SAR

[akupenguin]

Why so complicated? You do not need to recalculate SAR after cropping.
The 720x480 "16:9" has SAR=5760/4739 or 40/33 or 32/27 depending on how standards-compliant the DVD is.
The 720x352 still has SAR=5760/4739.

[SeeMoreDigital]

Those calculations are okay if you want to follow the ITU specification "to the letter". But if you want to follow a more logical approach, try the following methods...

If your movies aspect ratio (MAR) is 1.85:1, then this can be represented as a fraction of 37/20 (because 37 divided by 20 equals 1.85). Meaning it can be placed in an calculation, this way: -

37 352 13024 407
-- x --- = ----- lowest dominator --- giving a PAR of 407:450
20 720 14400 450

Which "could" be further simplifieid to say: -
708 68
--- = --
450 75


But like I mentioned before.... You need to know the "aspect ratio" of your source movie :)


Cheers

[Rash]

Right, OK. Thank you. ;)