Log in

View Full Version : Aspect Error - How much is acceptable?


Pages : [1] 2

Swan
18th April 2002, 15:01
Hello,

I have followed Doom9's Gknot how-to in every detail and I have read the FAQ. But I can't find this question answered:
How much Aspect Error is acceptable? I strive for 0.0%, of course, but there are times when neither lowering the resolution does the trick (it can be that the video becomes too small, height and width-wise) and also, changing the bitrate isn't always acceptable either.
How much Aspect Error is OK? When does it begin to show that the aspect ratio is bad?

skipper152
18th April 2002, 15:34
Hi,

I had the same problem as you. I read a while ago in everwicked's guide that .3% was acceptable. I just went to his site and he mentions 0% now.

However, what I do now is this

I start by putting the w-modul and h-modul at 16, and I adjust the size.
If I find the aspect ratio is off too much, I drop the H-Modul ONLY down to 8,then to 4, then to 2, until I get something that is close (if not exactly) 0% for the aspect ratio. I only remember having to drop it down to 4 at most.

With DivX 5.0 there was the problem with pink lines, et all. showing up if h-modul & w-modul weren't 16. But I think this has been fixed with 5.0.1 for the few movies I have done with 5.0.1.

I do believe the codec is optimized for h-modul & w-modul as 16, but if it takes a longer to encode (for me) it is not a big deal. I encode at night.

I'm not sure if this is the right thing to do, but it seems to be working for me.

skipper

Swan
18th April 2002, 16:05
Thanks, Skipper. I rarely use DivX 5, which I of course should have mentioned.

I will give your tips a try, unless you say it isn't valid for DivX 3.11 Alpha. :-)

Is there anywhere I can read more about the W- and H-modules?

skipper152
18th April 2002, 17:16
(Been kind of slow here at work, so I have a bunch of time on my hands to look at the forums....:0).

It's been quite a while since I used 3.11 (don't remember too much about it)...

I have several places that I like to visit on a daily basis (not in any order!)

http://www.doom9.net/
http://www.divx.com/
http://www.everwicked.com/

Just by browsing the forums, you can gain a tremendous amount of information! I use the search function quite a bit. If you have a question, probably someone already answered it.

I'm not sure if there is a lot to read about the h-modul and w-modul.
I have seen it in gordian knot, you can use just the calulator section on it. By setting the modul, the resolution changes by that amount (ie: setting it to 16 will be say 512->496->480). But it does help quite a bit when trying to determine aspect ratio and sizes.

I think aspect ratio should be valid for 3.11, however I'm not sure if it is optimized for the 16x16 as divx5 is supposed to be.

I'll take a guess that you are using gordian knot/nandub method....Just curious, why not switch to DivX 5?

skipper

Swan
18th April 2002, 17:42
Originally posted by skipper152
(Been kind of slow here at work, so I have a bunch of time on my hands to look at the forums....:0).


And I appreciate it a lot, Skipper. ;-)

[
http://www.doom9.net/
http://www.divx.com/
http://www.everwicked.com/
]

I check Doom9 most of the time. Everwicked; I'd never heard of, so thanks! :-)

amount (ie: setting it to 16 will be say 512->496->480). But it does help quite a bit when trying to determine aspect ratio and sizes.


A-haa. But there must be "invalid" or otherwise bad resolutions? Even if the aspect ratio is correct, the resolution could be bad, couldn't it? Sticking with the ones suggested in GKnot seemed safest to me. ;-)


I'll take a guess that you are using gordian knot/nandub method....Just curious, why not switch to DivX 5?


You are absolutely right. I also encode in Nandub alone and VirtualDub alone and use TMPGEnc. I am a video nerd. :-)
However, I have also started using the GKnot "combo".

I don't like DiVX 5 because the videos turn out kind of soft, not as sharp as with DivX 3. Also, when I play a DivX clip, and say, skip o another part in the video, the video starts speeding, but the audio is at the correct place. It's like the video playback is trying to catch up with audio, which is does after a while. I find this very annoying, I want the video and audio to start playing directly in sync, when I skip to another part in a video clip. Maybe this is a bug in DivX 5? Also, I have more problems with getting motion scenes (like a panorama shot of a beautiful scenery)100% smooth with DivX 5. With DivX 3, I never or rarely, get choppy motion in those scenes.

Thanks for all your help, Skipper!

skipper152
18th April 2002, 18:30
A-haa. But there must be "invalid" or otherwise bad resolutions? Even if the aspect ratio is correct, the resolution could be bad, couldn't it? Sticking with the ones suggested in GKnot seemed safest to me. ;-)

My guess is yes..I would believe that if you encode something at say 320xXXX and try to play it on a full screen 1280xXXX, then the player would have to stretch the picture. In DivX 5, for the same bitrate, a larger resolution may not end up looking good either. So there is a tradeoff. Most of my movies are done at 512xXXX.

I don't like DiVX 5 because the videos turn out kind of soft, not as sharp as with DivX 3. Also, when I play a DivX clip, and say, skip o another part in the video, the video starts speeding, but the audio is at the correct place. It's like the video playback is trying to catch up with audio, which is does after a while. I find this very annoying, I want the video and audio to start playing directly in sync, when I skip to another part in a video clip. Maybe this is a bug in DivX 5? Also, I have more problems with getting motion scenes (like a panorama shot of a beautiful scenery)100% smooth with DivX 5. With DivX 3, I never or rarely, get choppy motion in those scenes.

Is this true even if you use the sharp bicubic? Maybe without smoothers? The speeding I can understand, especially with b-frames (don't remember reading about a bug). The choppyness could be because the decoder was set at too high a level for the cpu...I think it defaults to 2 or 3 which might be too high.

I found that divx 5 gave some very nice results. I like my movies to look nice. I usually play them when I work the midnight shift on my PII 366 :D). So, probably if I played my movies on an HDTV, they might look crummy. I check them at home system after I make them and they look ok to me...Whatever works and looks good to you :D

Just some thoughts....

skipper

Swan
18th April 2002, 20:09
Originally posted by skipper152
[B]

My guess is yes..I would believe that if you encode something at say 320xXXX and try to play it on a full screen 1280xXXX, then the player would have to stretch the picture.


I never view my movies on the monitor (computer screen). I always output them to the TV. :-)

Is this true even if you use the sharp bicubic?

I have not tried that, unless it is set by default, but I certainly will. Thanks for the tip. :-)


The speeding I can understand, especially with b-frames (don't


It happens with *all* DivX 4 and 5 encoded videos I've seen, not just my own. And I've heard others complain about it as well. It does not happen to you? If you play a movie, then jump somewhere, does the video and audio play in sync right away? The video is not speeding?

The choppyness could be because the decoder was set at too high a level for the cpu...I think it defaults to 2 or 3 which might be too high.


I have an Athlon XP 1900+. Isn't it sufficient? ;-)
Or did I misunderstand you now?

might look crummy. I check them at home system after I make them and they look ok to me...Whatever works and looks good to you :D


Yes, to each his own.. Is that the correct saying? :-D
English isn't my native language.
May I also add that I hate spyware, which DivX 5 Pro is. :-)

/Swan

skipper152
18th April 2002, 20:52
I never view my movies on the monitor (computer screen). I always output them to the TV. :-)

We're in opposite ends on this one; I have never watched my divx movies on the TV. Too lazy to run the wires from one end of the house to the other.:D

The speeding I have never seen happen on divx 4 (even when I had my Athlon 600). It has happened on divx 5. But when I watch the movie, I put the movie in, press the play button, then sit back with a bag of popcorn....I rarely do searches or skips in the movie, so for the second it takes to catch up doesn't bother me. I spot check the movie right after it is made, then watch the movie from hard drive before I burn it onto a CD.

I have an Athlon XP 1900+. Isn't it sufficient? ;-)
I am using an Athlon 1.4GHz (not an XP). I believe the 1900+ is equivalent to a 1.67GHz. The decoder I have set at minimum. There are other things too. If you are running from the hard drive, the drive could be fragmented. Using a CD-ROM DMA might not be turned on. Other processes running in the background? (antivirus). I have read that even a mid setting for the decoder would make some of the high end systems stutter once in a while. Just remembering what I read in other places. But in general, I would think your system would be more than sufficient.

As far as spyware, I think you are right. I am not always connected to the internet (in fact rarely), but some people are, and not want their personal information transmitted to wherever...

Don't know if this helped out anymore, but some more information....

skipper

English isn't my native language
English is mine, but I still have trouble with it....:eek:

Swan
18th April 2002, 22:23
Originally posted by skipper152
We're in opposite ends on this one; I have never watched my divx movies on the TV. Too lazy to run the wires from one end of the house to the other.:D


:-D
I don't think it's enjoyable to watch them on the computer monitor. The gamma is all wrong and my gtraphics card doesn't allow me to set the gamma to that of a TV. Perhaps yurs does?


The speeding I have never seen happen on divx 4 (even when I had my Athlon 600). It has happened on divx 5. But when I watch the movie, I

I have a very trustworthy friend (meaning technically advanced and excellent at video "matters" and he says he's got the speeding problem with both 4 and 5, just like me. ;-)
Until I find a way of making it disappear, or it is fixed by the authors if it indeed is a bug, I can't accept using DivX for the stuff I encode (normally TV shows and such, which I capture).



There are other things too. If you are running from the hard drive, the drive could be fragmented. Using a CD-ROM DMA might not be turned on. Other processes running in the background? (antivirus).

I defragment once a week, I defintely have DMA enabled and I *hate* the "monitoring" or "always scanning" feature of antivirus software. I have not enabled it, have not even installed it, and never would. I scan files "by hand"/"on demand". Always. ;-)


what I read in other places. But in general, I would think your system would be more than sufficient.

I think so too.


Don't know if this helped out anymore, but some more information....

It's always interesting to discuss a common interest with someone knowledgeable, Skipper. Thank you for the tips and everything.


English is mine, but I still have trouble with it....:eek: [/B]
That's kind of you to say. ;-)

/Swan

TheWEF
19th April 2002, 00:05
Originally posted by skipper152
I start by putting the w-modul and h-modul at 16, and I adjust the size.
If I find the aspect ratio is off too much, I drop the H-Modul ONLY down to 8,then to 4, then to 2, until I get something that is close (if not exactly) 0% for the aspect ratio. I only remember having to drop it down to 4 at most.

do NOT set H- or W- module lower than 16!!!!!
(divx is encoded in macroblocks of 16x16 pixels).

simply check "smartcrop all" to get 0% ar-error.

:rolleyes:

wef.

InfoCynic
19th April 2002, 03:12
I hate to disagree with someone as prominent as the TheWEF, but oh well. :)

Smart crop all can SERIOUSLY kill a ton of pixels. You ever watch how many pixels that setting will chop off? I've seen it take like 10 from multiple sides if I have a really bad resolution set.

I used to use Smart Crop all the time... but then I got to thinking... if I look up the movie's AR on imdb, or know what it is for TV (4:3 or 16:9 depending on if the show is WS), wouldn't it be better to crop until all the black was exactly gone, and then find a good resolution to encode at? If I have to use a lower resolution to stay at .2 bits/(pixel*frame) AND keep my AR, then that's life. But I often found that a 1.85:1 movie might be 1.9 or so when I've cropped it completely, before resizing. Now in reality, that's not a horrible % error, but if I start using Smart Crop, it might actually get worse, not better, as I adjust my resolution.

So now my method is like I said: manually crop as close as you can and then find a resolution to work with. I only started this method recently (since Divx 5.01), but I've been keeping track of the resolutions I use for different ARs. Format is AR -> WxH (W-modul, H-modul). I don't consider W- or H-modul 2 valid because I use Divx 5.01 and it requires 4 (but still recommends 16).

4:3 (1.333:1) -> 512x384 (32, 32), 576x432 (32, 16), 640x480 (32, 32)
16:9 (1.778:1) -> 512x288 (32, 32)
1.85:1 -> 608x328 (4,8)
2.35:1 -> 640x272 (32,16), 564x240 (4,16)

So perhaps TheWEF or someone who agrees with him would care to explain what exactly is so horrible about not using the 16-moduls?

skipper152
19th April 2002, 15:21
I hate to disagree with TheWEF also, but I'm sure TheWEF has some inside information that we don't know about.

SmartCrop does not always work properly. I believe I saw in everwickeds guide the Width should not be less than 704 (haven't really looked through the guide in a while). However, I can think of one movie that this does happen (Rocky Horror Picture Show).

I believe I do the same as InfoCynic. Manually crop the black bars, then find a good width, then adjust the Height to as close to the aspect ratio as possible, starting at 16, going down, but not using 2.

The reason I do it this way is in 5.0.1, this problem was supposed to be fixed (taken from http://www.divx.com/divx/divx_win_versions.php). I did not do this with 5.0

Fixed an issue where encoding content with resolutions not divisible by 16 would cause purple and pink artifacts to appear in the encoded content. These artifacts no longer appear. You can now encode content that has resolutions divisible by 4 or 8, however we still strongly recommend encoding only content with resolutions divisible by 16, as the 4 and 8 dimensions will not encode as efficiently. (There remains a known issue about resolutions divisible by 2.)

Not encode as efficiently I took to mean, will take longer (which is not a problem by me). Not that it wouldn't encode.

So please enlighten us someone.

Thanks

By the way TheWEF, I know you have heard it many times, but it is a great program!

skipper

Mac Sidewinder
19th April 2002, 16:39
<<SmartCrop does not always work properly. I believe I saw in everwickeds guide the Width should not be less than 704 (haven't really looked through the guide in a while). However, I can think of one movie that this does happen (Rocky Horror Picture Show). >>

Why not go below 704? (besides the obvious reason of losing quite a bit of info from each side)

<<Not encode as efficiently I took to mean, will take longer (which is not a problem by me). Not that it wouldn't encode. >>

Take longer or not compress as much making your file sizes larger?

Mac

InfoCynic
19th April 2002, 17:07
I try not to go below 512, but I either use the bits/(pixel*frame) or a compressibility check to determine what resolution to use. I try not to go above 640 either, so nobody ever has to view it at less than 100% magnification (some TV-out cards use 640x480), but it's rare that I'd have enough space to compress well at 640 anyway.

And I wasn't very convinced that everwicked would say never to go less than 704, so I check his guide, and it doesn't say anything like that. Here's what it does say (http://www.everwicked.com/content/DivX_5_Guide/divx5-05.php):

"Select 16 as the W and H Modul. This will make sure that you resolution is a multiple of 16. It will also give you a variety of resolutions. Slide the bar from the right to the left until you reach a Bits/Pixel*Frame value close to 0.190.

Important: Do not go lower than 0.175 in any case. Always check that any of the W-Zoom, H-Zoom values do not exceed 100%. Also make sure that the Aspect Ratio error is 0.0%."

skipper152
19th April 2002, 17:59
Take longer or not compress as much making your file sizes larger?
A definite possibilty. But for the DivX 5.xx movies I have done (granted not a lot on DivX 5), the file sizes seem about right, and the quality has been fine to me (yes, I know, this is subjective...)
And I wasn't very convinced that everwicked would say never to go less than 704, so I check his guide, and it doesn't say anything like that. Here's what it does say
It has been a while since I looked at the guides, but it is not there now after giving it a quick look, so I guess this is not a valid point anymore.

Sorry to keep editing, but I just did a quick search on everwickeds site and found a link to support the 704 item (at least that I did read it, and I am not totally insane, just a little crazy)http://www.everwicked.com/forums/showthread.php?s=&threadid=511&highlight=width
but it was for DivX 4, not 5, and had to do with some codecs which I wasn't using at the time, I guess.

I do apologize for getting my "facts" mixed up...

Why not go below 704? (besides the obvious reason of losing quite a bit of info from each side)
Seems to be after taking a quick look at the link, it had to do with DivX 4, flipping video. Not sure if I fully understand it. But there is a link if you wish to check it.

Important: Do not go lower than 0.175 in any case. Always check that any of the W-Zoom, H-Zoom values do not exceed 100%. Also make sure that the Aspect Ratio error is 0.0%."

So if I can ask a question, if no one wants to shoot me yet for my memory possibly failing (not the computers, my brain :D ), and writing a short novel here....

With the above quote, how do we ensure the apect ratio is 0.0% if we are to maintain the 16 modul (as TheWEF mentioned)?

Thanks,
skipper

TheWEF
19th April 2002, 21:55
Originally posted by skipper152
I hate to disagree with TheWEF also, but I'm sure TheWEF has some inside information that we don't know about.

here you go: if you don't resize mod 16/16 it doesn't even make sense to crop. keep the black - same result.

who is everwicked? ;)

wef.

InfoCynic
19th April 2002, 22:18
This conversation should almost be moved to Divx5 encoding, but for now...

I'm sorry, but you'll have to EXPLAIN more than that... I can resize to a width/height of modul-4 after cropping... and then I have a non-letterboxed divx as a perfect aspect ratio that I didn't overcrop. Here's an example:

I have a PAL clip that's 16:9. I cropped 74 pixels off top and 72 from the bottom and none off left-right. This shows an AR of 1.786, which is reasonably close to 1.778. If I slide the width over to 560 (mod-16), the height at mod-16 is 320. Without activating Smart Crop, the AR is 1.750. Now that's STILL not horrible. That's about 1.5%. I activate Smart Crop All, and the cropping settings jump to 80 off top and 79 off bottom. that's 11 vertical pixels I've lost just like that. To me, that's unaccetable. WORSE, the AR is now 1.842! That's 3.6% error. Of course, it SAYS it's 0.0% error, but it's comparing your resized AR to your cropped AR, USING THE SMART CROP VALUES. I'm taking (abs (ResizedAR - TrueAR))/(TrueAR) where abs is the absolute value function, Resized AR is the value given in the AR field next to H-modul in GKnot, and TrueAR is the known AR of the source media (in this case, using 1.778).

Now, if you use an h-modul 4 or 8... you can use 312 for the H. This gets you an AR of 1.795 (error: 0.96%).

So... explain again why I should use an H-modul of 16 and smart cropping? I'm not trying to be insulting or demeaning, I just don't see why to use your method, if encoding speed doesn't matter, and even if it DID, why should I sacrifice 11 vertical pixels for example?

TheWEF
20th April 2002, 00:39
discussions about AR are so boring for me, it almost hurts. i had them about 1500 posts ago. so please don't be angry with me if this is my last post in this thread. :)

but i will try to explain some things (although i don't HAVE to ;) ).

1) what you read on the box is NOT TrueAR, it's ReadOnTheBoxAR, nothing more (mostly you just read the name of the film AR, but you might be surprised that so called 2.35 really is 2.40, then on the dvd it is cropped..., or not..., well...)!

2) dvd AR and resizing is a more complex matter than it seems at first glance, because computer software dvd players do not resize the same way as standalone players do. (maybe search the forum for ITU standard to find out more).

3) 100% correct AR is something you can never be sure of unless you photographed the movie and mastered the dvd yourself.
but in most cases you can trust gknot. if the dvd was mastered following the standard rules, gknot shows 0% error and ITU is checked (in options), this is the best AR you can get.

4) i do NOT think that 0% aspect error is something you have to aim for. actually nobody can tell if the AR is 5% off. a badly adjusted monitor or tv can produce bigger errors than that. the problem is that 5%(mastering) + 5%(resizing) + 5%(display) can be 15%...
so the question to ask yourself is not "0% error or not?" but "+3% or -3% error?".

5) why only use mod 16/16?
first resolution is 640x272 pixels = 40x17=680 divx macroblocks.
second is 640x276 (mod 4) pixels = 40x18=720 blocks.
second has 1.47% more pixels but 5.88% more blocks.
both encoded at the same quality the second file will be more than 1.47% bigger in filesize than the first one (not 5.88% but somewhere in between). so you are using filesize on pixels that you cannot see.
in other words: the 16/16 file will be of better quality at the same filesize.

/me exhaused :p

wef.

skipper152
20th April 2002, 02:09
@TheWEF
Thank You. I wasn't getting angry, just disappointed. Now I know it is my filesize to waste, and I might not be getting the best quality that I can.

By the way: http://www.everwicked.com/
Check out the site sometime, it's pretty good :D

(And as a side note, it seems to me that people in these forums are ready to jump all over someone, but when that someone corrects themselves and apologizes, shows links, asks questions, they are ignored; maybe it is because I only have 7 posts or I'm just a lowly programmer/analyst. Just an observation so far.)

Now:
http://www.everwicked.com/content/DivX_5_Guide/divx5-05.php
Go to the Options tab. Uncheck the Follow ITU-R BT.601 Standard option.
To check or not to check.....

Time to go back an re-encode some movies to see if there is any difference (and I'll try to stay out of the forums from now on).

Thanks

TheWEF
20th April 2002, 02:45
Originally posted by skipper152
@TheWEF
Thank You

you're welcome.

Originally posted by skipper152
And as a side note, it seems to me that people in these forums are ready to jump all over someone, but when that someone corrects themselves and apologizes, shows links, asks questions, they are ignored; maybe it is because I only have 7 posts or I'm just a lowly programmer/analyst. Just an observation so far.

are you talking about me? this thread? another thread?
please tell me where you've been jumped all over or ignored? :)

Originally posted by skipper152
and I'll try to stay out of the forums from now on

hey, hey! i just spent a lot of time on you!
and you did learn a lot! :p
i think you have to stay and take care of the next newbie that comes along. ;)

what's going on in here, i don't get it any more... :confused:

wef.

skipper152
20th April 2002, 05:24
I don't want to turn this into a one-on-one or personal post, but since you posted and are asking questions, you, of course deserve a response.
</BEGIN LONG RANT>
are you talking about me? this thread? another thread?
Definitely not you!
We, as posters, have to realize that if it takes a couple of days for the folks who write these great programs to get back to us, we need to understand that. After all, if you folks spent all of your time in the forums, we wouldn't get these great programs.

By the way, if you are reading this post, check out the pie thread.
http://forum.doom9.org/showthread.php?s=&threadid=14432
Yes, I read the stickies too.

I can sympathize with you a little more than the average person. I know what it is like to write code. Tracing a problem can take days, getting requirements, changes from divx, regression testing, updates; and of course you have work/school, friends-family/wife-kids, moderator in this forum, probably get friends/collegues picking at your brain for information, assistance in testing, helping them debug. Oh, I forgot, sleep/eat. Gosh, when do you have time for a beer?

Anyway, if it hasn't been said enough, thanks to you and all the folks who give their time for the programs we use.

Maybe "jump" was a little too harsh. Mac & InfoCynic both questioned what I wrote on Apr 19, at 10:21. I took the time at 12:59 to write my "mini-novel". InfoCynic has been on to question you, but didn't give my post one mention that he even read it. Maybe I missed it. I'm not crying about it, but I would like to know if it is common practice to question someone and ignore the response after they have taken the time for it? I can go for another cup of coffee rather than respond. Maybe I have a lot more to learn about the forums (forget about that I have a lot to learn about divx) than I think I do.

hey, hey! i just spent a lot of time on you!
Yes, you did (sorry you missed out on that beer :p )
And I did learn a bunch; I have been surfing 3 different DivX forums for about 2 years now. I recently joined from what I read in one of the forums (I think it was everwickeds; guess I've been spending too much time on his site :D );
The experts train the current newbies. These newbies then become experts, which train new newbies. I've been getting information out of these sites, I wanted to try to return some.
But if I am already "branded" as a person not knowing anything (hence being ignored) after about 9 posts, then I shouldn't be wasting the bandwidth with responses. Granted, this is probably an isolated incident (also, some people can't spend all day on the web) and the reason why I said "Just an observation so far". I'll have to see how it goes.

(Man, every time I post, I write a mega novel here, gotta learn how to type in shorthand ;) )
</END LONG RANT>

Gee wef, I didn't know you cared so much...:D

Thanks

TheWEF
20th April 2002, 05:57
lol - rant?! - just compliments! :)

i think you are exactly the kind of person we need in this place! please stay and post, just grow a little thicker skin.

mac and ifocynic, i don't know you, but be nice to him! ;)

(and if they are not you shouldn't care too much...)

wef.

InfoCynic
20th April 2002, 06:56
Hm. Some interesting things to consider. I'll try to remember them in the future. I guess I can stick to 16x16. :) Thanks for the help. What other major crises can we solve? :D

Skipper, I did read your posts, but I have a standing policy: If I don't know the answer to a question, I don't answer it. :) And my posts tend to ramble as-is without bringing up multiple things (although I believe there was some discussion of post-processing earlier, and I can say that with an Athlon 750 and 256 MB of system RAM and a 32 MB ATI Rage Fury Pro, I was able to use FULL post-processing on full-screen divx4... i upgraded heavily before divx5 came out, so it would be less meaningful. And as far as the "skipping" problem, this comes from not seeking to a keyframe AFAIK. For example, if you use chapterXtractor to setup MicroDVD chapters and you just use the exact frames, you'll see that problem, but if you synch it to an AVI file, it will use the nearest keyframe, and then you don't have the problem of video lagging way behind audio for a few seconds). See? Ramble. :D

I just happened to be browsing the forum and since a thread on AR seemed interesting, I popped in. :) And since I have a very dull job, when I got the e-mail alert that I had new posts, I was back here within a minute or two usually. :)

How 'bout a theoretical question: is there ANY time when using non-mod 16 would make more sense? (And for the smart-aleck out there who is going to post a one-word reply, :p ).

And I'm nice to everyone. Almost. (Except ppl who say "i'm right, you're wrong" without listening to what you have to say....)

skipper152
20th April 2002, 13:43
i think you are exactly the kind of person we need in this place! please stay and post, just grow a little thicker skin
Thanks WEF; that's quite a compliment coming from someone as respected as you.

If I need to grow a thicker skin, then "make it so..." :).

@Info
Wasn't looking for an answer if you didn't know it, just an acknowledgement that the quote and link were read. We're all human (I think), and make mistakes once in a while. Hey, I read about the 704 problem, just in the wrong forum which didn't apply here :D.

To get back on track....

I attempted using the 16 for both H & W on one of the movies I had done. This time, I checked the ITU-R Standard in the options (probably not a good idea to change 2 elements at the same time; not very scientific. Oh well....)
However, it says in everwickeds guide (on the 5th page) to uncheck it.
Let autocrop run. I put back the pixels it took off on the sides (sorry, I don't like missing the edges, bothers me...)
Aspect was off (not much 1.2%, I think). So, what I did was put back a little of the black borders on the top/bottom to get the AR to 0.0%. It was only a couple of pixels or so...
Did the encoding, muxed, noticed the file was about 30 MB larger than before. Hmmm? Well, I accidentally used sharp bicubic, rather than neutral bicubic, which I normally use. So, we use up a few more CPU cycles and try again.
So far, I have not noticed anything in the speed department (Athlon 1.4GHz / 512MB Ram), but the system usually runs at 100% CPU during encoding.
Now, I should mention this is about a 2 1/2 hour movie, and Gordian comes back with it being very compressible (around 87%). I can agree with this as it is very slow motion and dark. Don't know if the info is going to make any difference. But the bitrate Gordian gives is 517 and I am using 640, so we will see (aiming for a 700MB file total).
I will leave the system running while I take the kids to soccer and hopefully get back later in the afternoon (depends on where you are!) with more info after it is muxed. But I'm not sure if I will be able to tell a quality difference.

skipper

Swan
20th April 2002, 14:17
I started this thread and I appreciate the WEF's (and others) comments a lot.
However, I don't feel I got my question answered yet. ;-)

How much of the GordianKnot tool's reported aspect error is "OK", meaning the human eye cannot perceive it?
Is -/+0.1% OK?
What I'm trying to find out is; if I cannot achieve 0.0% without severly cropping the video, is it OK to accept a little Aspect Error? If so, how much?

I have always followed the rules in Doom9's guides and have not changed the W- and H-modules to anything but what he suggests.
I strive for 0.0% Aspect Error, of course. But as I encode more and more (DivX3), I see that I'm having problems getting 0.0% Aspect Error in some cases. I just want to do this properly. :-)

Razor04
20th April 2002, 15:27
Hi! TheWef mentioned this in a previous post, you probably just missed it. Please look at points 3 and 4 specifically. For me personally, as long as the error is under 3% it is something that I will never notice. Once the movie is encoded I don't go back and compare it to the original, and really it isn't noticable. Like TheWef said, the monitor could technically introduce more error. For instance, think of 1280x1024 that isn't a 1.33 aspect ratio like every monitor out there, but is instead is a 1.25 aspect ratio. If you are running in that mode you would introduce some error anyways. Like he mentioned, it is really how much error you can tolerate before it becomes noticable to you. Hope this helps.

Originally posted by TheWEF
discussions about AR are so boring for me, it almost hurts. i had them about 1500 posts ago. so please don't be angry with me if this is my last post in this thread. :)

but i will try to explain some things (although i don't HAVE to ;) ).

1) what you read on the box is NOT TrueAR, it's ReadOnTheBoxAR, nothing more (mostly you just read the name of the film AR, but you might be surprised that so called 2.35 really is 2.40, then on the dvd it is cropped..., or not..., well...)!

2) dvd AR and resizing is a more complex matter than it seems at first glance, because computer software dvd players do not resize the same way as standalone players do. (maybe search the forum for ITU standard to find out more).

3) 100% correct AR is something you can never be sure of unless you photographed the movie and mastered the dvd yourself.
but in most cases you can trust gknot. if the dvd was mastered following the standard rules, gknot shows 0% error and ITU is checked (in options), this is the best AR you can get.

4) i do NOT think that 0% aspect error is something you have to aim for. actually nobody can tell if the AR is 5% off. a badly adjusted monitor or tv can produce bigger errors than that. the problem is that 5%(mastering) + 5%(resizing) + 5%(display) can be 15%...
so the question to ask yourself is not "0% error or not?" but "+3% or -3% error?".

/me exhaused :p

wef.

skipper152
21st April 2002, 03:16
OK, some info as promised....

(See my post above for what I did....)

The resultant filesize came out about 6MB larger. This is not earth shattering, however interesting....

I'm guessing that because the movie was very compressible to begin with, there is very little additional compression that the codec has done. However, someone with a better eye than mine might see it.

The 6MB filesize is the additional data WEF was taking about. Rather than using the filesize on data we cannot see, we are using the filesize for something we can see and it does contain valid data (+ a little more).

I am trying another movie which I had some difficulty with and I acutally saw some problems. I will see if those problems will be going away...

skipper

TheWEF
21st April 2002, 05:24
are you encoding 1-pass quality based or 2-pass?
divx5?

(if you are doing 2-pass it doesn't make sense to compare filesize)

what exactly are you trying to proove?

wef.

TheWEF
21st April 2002, 05:33
maybe i should say it that way:

IMHO (and you should trust me ;) ) using anything lower than mod 16/16 or keeping the black bars is NOT an option.

either accept the ar-error or use smartcrop.

skipper152
22nd April 2002, 18:43
IMHO (and you should trust me ) using anything lower than mod 16/16 or keeping the black bars is NOT an option.
Irrelevant of my testing, I was planning on trusting you anyway, and do what you recommend. ;).
I'm ripping my last movies (have to go buy some more), so I thought I would do some testing. After all, what's the sense in having a computer if you can't run it 24x7 :D. I'll deal with the aspect error.

I should tell you I only recently started playing around with non mod-16 (these last movies I have on my HD). Simply because everwickeds guide mentioned to make sure of a 0.0% aspect ratio, and the DivX 5 updates said the other mods were fixed, but recommended to stay at 16.

what exactly are you trying to proove?
Gee some long winded person ;) a few posts back said
in other words: the 16/16 file will be of better quality at the same filesize.
So, I thought I have some time on my hands, my last movies are ripped to the HD, lets try it. Can I see the difference? I had my W-mod at 16, however my H-mod was at 8. I knew what that looked like.

So, I set up a 16/16 with the same bitrate. I expected the filesize to come out the same size.
However, I later realized 2 problems I had with this:
1) I remember reading (somewhere) that the pixels where the black border and the movie meet gives the codec problems (doesn't encode as well).
2) That once I change the output resolution to accommodate the new 16/16 or 16/8, this changes the bits/pixel which gives one test an unfair advantage over the other if we are looking at quality.

So, either one (or both) of these could account for the 6MB increase.

(if you are doing 2-pass it doesn't make sense to compare filesize)
Well, I am doing 2-pass and using Divx 5. :( If you are telling me this, then

#ifdef DIVX_TALK
#define skipper confused
#endif

Since my understanding (maybe incorrect) that at a given bitrate, given the same filters used, the resolution used would not matter to the filesize, but might change the quality of the movie (to a certain point). This is because the bits/pixel would change.

So at the point I did the test: change the resolution to 16/16, keep the same bitrate. I thought the filesize should not change, however when it did, I thought it was the additional data from not doing 16/16 but 16/8 from before. In the example you gave WEF, I think you would be wasting about 75% of 40 macroblocks.

skipper

TheWEF
22nd April 2002, 21:38
one main purpose of two pass encoding is to reach a certain filesize (= a certain average bitrate), usually to make a movie fit on one or two cds.
(second purpose is optimized "bitrate"-distribution within that certain filesize)
divx3 never had 2-pass functionality, you never knew what filesize you ended up with (regular) 1-pass divx3.
but i guy called nando aka _nn_ invented a very fine method for 2-pass encoding with divx3, he wrote a tool for it ("nandub", based on vdub sources) and called the method sbc (smart bitrate control).
it is so perfectly implemented in his tool that it is possible to keep the filesize deviation within a few kb.

unfortunately divx5's 2-pass-implementation is not that good. as you found out already, deviation of +/- 10mb in nothing unusual.
but don't even think about what causes these deviations, you will not find any pattern. it's more or less a question of luck.

this is the main reason why i still don't like divx4/5.

so this is why it doesn't make sense to compare the filesize of different 2-pass-encodings.

wef.

Dark-Cracker
22nd April 2002, 23:47
Hi , sorry but my english is very poor and i dont have understand very well how to make for getting an Aspect Ratio Error near to 0.0%

i am trying to modify a tool made by (DrMAD / bloodlust) and i try to add a smart crop all function, this tool crop automaticly the black border, then choose an aspect ratio (2,35/1.85/1.77/1.33....) and it resize the picture

Perhaps some can to help :
- first question how to find if a .d2v is PAL or NTSC ? (perhaps with the framerate)
- can someone try to explain me how to calculate the AR and the AR error ? (with an expemple)

- can someone try this version of the tool (very buggy, and in french) and try to explain me if i should add a "smartcropall" option or if it work nice without ?

http://membres.lycos.fr/darkangel47/AVSToolBeta.zip


thank u for helping me :-)

PS: sorry for my crap english, it's not my mother language.

TheWEF
23rd April 2002, 03:58
Originally posted by Dark-Cracker
- first question how to find if a .d2v is PAL or NTSC ? (perhaps with the framerate)

resolution and framerate is possible.

pal: 720x576, 25fps
ntsc: 720x480, 29.970fps

Originally posted by Dark-Cracker
...this tool crop automaticly the black border, then choose an aspect ratio (2,35/1.85/1.77/1.33....) and it resize the picture...
- can someone try to explain me how to calculate the AR and the AR error ? (with an expemple)

do NOT choose an AR (you could choose anything - what is correct...?)

the only thing you have to decide manually is whether the source is anamorphic or not.

the rest works like this (have a look at gknot):

pixel aspect ratios:

official-ITU-standard:

PAL-anamorphic:1.457
PAL-nonanam.:1.093
NTSC-anam.:1.212
NTSC-nonanam.:0.909


SW-DVD-player-standard:

PAL-anamorphic:1.422
PAL-nonanam.:1.067
NTSC-anam.:1.185
NTSC-nonanam.:0.889


if AspectMode=Anamorphic then
begin
if cbITU.Checked then
begin
if AspectMode=PAL then ResCalc.PixelAspIn:=(59/54)*(4/3)
else ResCalc.PixelAspIn:=(10/11)*(4/3);
end
else
begin
if AspectMode=PAL then ResCalc.PixelAspIn:= (16/9)*(576/720)
else ResCalc.PixelAspIn:= (16/9)*(480/720);
end;
UpdateAll;
end;

if AspectMode=NonAnamorphic then
begin
if cbITU.Checked then
begin
if AspectMode=PAL then ResCalc.PixelAspIn:= 59/54
else ResCalc.PixelAspIn:= 10/11;
end
else
begin
if AspectMode=PAL then ResCalc.PixelAspIn:= (4/3)*(576/720)
else ResCalc.PixelAspIn:= (4/3)*(480/720);
end;
UpdateAll;
end;


now use your brain and figure out the rest yourself! :D

wef.

Dark-Cracker
23rd April 2002, 11:23
thank u i will try to modify the prog with this new informations.

a very big greet to u, happy to see there is alway someon for giving a hand :-)

everwicked
24th April 2002, 09:25
Originally posted by skipper152

Sorry to keep editing, but I just did a quick search on everwickeds site and found a link to support the 704 item (at least that I did read it, and I am not totally insane, just a little crazy)http://www.everwicked.com/forums/showthread.php?s=&threadid=511&highlight=width
but it was for DivX 4, not 5, and had to do with some codecs which I wasn't using at the time, I guess.

Seems to be after taking a quick look at the link, it had to do with DivX 4, flipping video. Not sure if I fully understand it. But there is a link if you wish to check it.

Thanks,
skipper

It was indeed there in previous versions. I was reffering to the cropping width and not the output.

skipper152
24th April 2002, 15:10
Thanks everwicked. I didn't think I was going totally crazy (yet :)).
When I originally saw that you posted here, I thought 'Now I'm really off base...':D

Taken from mac's post:
<<SmartCrop does not always work properly. I believe I saw in everwickeds guide the Width should not be less than 704 (haven't really looked through the guide in a while). However, I can think of one movie that this does happen (Rocky Horror Picture Show). >>

Why not go below 704? (besides the obvious reason of losing quite a bit of info from each side)

Since we were talking about smartcrop, I thought it was kind of obvious that it was for cropping, not for the output resolution. But if I was not clear enough in my post, I am sorry.

Thanks again,
skipper

everwicked
24th April 2002, 17:30
No need to be sorry mate, I just clarified the situation.

Sharro
29th April 2002, 16:19
I have had movies with black bar on the left...none on the right...bars on left and right... usually top and bottom are a rule.

So...I do my cropping manualy. If u guys play around...you will always find an optimized cropping for 0.0% aspect error.

And I believe the difference on the location of the black bars...might affect smart crop...but that has to be TheWef to confirm it.

Take care.

Sharro

Swan
10th May 2002, 20:49
I always go for 0.0% Aspect Error. However, I have yet to find a movie where getting there didn't mean being forced to cutting a few pixels into the movie (not just the black areas). Is this normal?

InfoCynic
10th May 2002, 20:56
As I have said before, the "0.0%" shown by GKnot is misleading, since it calculates the error based on your Smart Crop values and not on your Pixel Crop values (you'll see it change when you activate S. Crop).

But that notwithstanding, it's almost guaranteed to have to crop into a movie a few pixels to get that 0.0% error. Personally, I look for the resolution that doesn't require as much overcropping and use that. In theory, that will be closer to a 0.0% error anyway since WEF says that the cropped AR is the correct AR, regardless of what the box or IMDB says.

Swan
10th May 2002, 21:04
Originally posted by InfoCynic
As I have said before, the "0.0%" shown by GKnot is misleading, since it calculates the error based on your Smart Crop values and not on your Pixel Crop values (you'll see it change when you activate S. Crop).

So, if I want to use the Pixel crop option, cropping myself, how can I make sure I will get the correct Apect Ratio?

Even when using Smart Crop, I often get an Aspect Error of say 1.2%, until I resize. Only then do I get Aspect Error 0.0%. Is this how it should be?

I have experimented with first using the Pixel crop, then pressing Smart Crop. Sometimes Smart Crops changes the amount of cropped pixels, but not always.

I have also encoded movies using just Pixel crop, then setting the resolution (getting, what I know understand was a misleading 0.0% Aspect Error). This has also meant cutting a few pixels into the movie, but not as many as with Smart Crop. Is this method totally wrong?

Swan
10th May 2002, 21:09
Sorry. I meant I have first used Pixel Crop to get it to look as I want, then selected "Smart Crop All".
I get Aspect Errors, until I resize, with Auto Crop.

InfoCynic
10th May 2002, 22:13
There are many different schools of thought on this matter. Let me explain how I do it:

1. Ignore GKnot's AR erorr %. It's bunk. :) (But we still love you, TheWEF)
2. using some combo of Auto and Pixel crop, crop the movie exactly or as close as your eyes can tell.
3. Compare the AR to the AR shown in the output area by the AR Error %. You want those numbers as close as possible. Resize until they are (but watch the bits/pixel*frame or the compressibility check % so you maintain quality). If you want to calcualte the error, it's

(CroppedAR - ResizedAR) / (CroppedAR)

4. You can smart crop now and that should get you even closer without cropping too much of your movie off (but you don't have to smart crop if you don't want to... but then you'll have a slight AR error).

TheWEF
11th May 2002, 00:10
Originally posted by InfoCynic
Ignore GKnot's AR erorr %. It's bunk....
If you want to calcualte the error, it's
(CroppedAR - ResizedAR) / (CroppedAR)

well, not more or less bunk than your formula.
this is how i calculate the error:
100*(CroppedAR / ResizedAR)-100

Originally posted by InfoCynic
You want those numbers as close as possible.

i think you can see that in both cases result is
Error = 0 if CroppedAR = ResizedAR

again i want to point out that Error=0 is a theortical thing, e.g. depending on whether ITU is checked or not.
there is NO WAY to make sure you really have correct AR.

for detailed information read this (http://toolbox.sgi.com/TasteOfDT/documents/video/lurker/pixelaspect.html). please notice the end of the last sentence:
...use your own judgement.

IMHO it is absolute nonsens to decide framesize based on AR-error.
check compressibility and quality, then resize!

wef.

TheWEF
11th May 2002, 00:32
Originally posted by InfoCynic
As I have said before, the "0.0%" shown by GKnot is misleading, since it calculates the error based on your Smart Crop values and not on your Pixel Crop values (you'll see it change when you activate S. Crop).

i don't understand this statement?
of course the error value changes if you change cropping (if you crop more - that's what smartcrop does)

...because CroppedAR changes - the part to resize gets a different height or width.

wef.

MaTTeR
11th May 2002, 00:48
Originally posted by TheWEF

IMHO it is absolute nonsens to decide framesize based on AR-error.
check compressibility and quality, then resize!

Been following this thread for quite awhile. I have to totally agree with the above statement. It seems a waste of time IMO. You guys must have better eyes to see these AR errors than 99.5% of the rest of the living population:)

skipper152
11th May 2002, 02:58
Wow, I can't believe this thread is still going on. Must be a really hot topic :D

I don't believe the AR Error % is really bunk, but rather an indicator of a best guestimation of how far off the aspect ratio is. If I start seeing this +/- 5%, I might start to question my input source or something I have done.

there is NO WAY to make sure you really have correct AR
Well, why? Maybe because....

Are you watching the movie on a pc monitor or TV? Will you only be watching this movie on TV or pc or can you switch back and forth? Does it look different?

Is the source ITU compliant or not? How can you tell?

Are you watching on a standalone player or through some software? Are you only using software or can you switch? Today I use software, tomorrow I want to use a standalone player..Is the aspect ratio incorrect now?



Could the 1.2% aspect error you are getting actually be the correct aspect ratio, but getting it to 0.0% is an incorrect aspect ratio?

I've gone back and done a couple of movies as wef suggested (yes, using mod-16 :D. I've been in this thread before). I have had an aspect error of 2.6%. And it looks fine (to me). I take a quick look through the movie, find a section with an object that I know (a globe, a clock, car tire, etc...), encode this section (say about 5 min worth), then look at it. If my globe looks like an oval, then I go back and try again.

Maybe the topic of this thread should be
"Aspect Error - How much is acceptable to me"

I may still have this wrong, but I would go with what looks good to me even if the aspect error doesn't say 0.0%.

skipper

Swan
11th May 2002, 19:14
Originally posted by skipper152
Maybe the topic of this thread should be
"Aspect Error - How much is acceptable to me"


Or rather "How should I crop and resize to get the least possible
Aspect Ratio Error".
I have only followed the GKnotguide:

Phase 1 - Using gknot to crop, choose resolution and create AVS file in preparation for first nandub pass.

(steps 1-3 removed)

4> change to the resolution tab and choose the "pixel" radio button under the crop settings, while watching the d2v viewing
window, use the crop-buttons to remove black borders on all for sides.
5> In the d2v viewing window select "resized" in the "view" menu.
6> now watch the video and manually set 'non anamorphic 4:3' or 'anamorphic 16:9' so that the pic is not stretched or
squeezed.
7> (optional) change the "pixel" radio button to "smart crop all".- this will give 100% correct AR but crops off additional
pixels...

If the Aspect ratio Error box is practically useless, as you all seem to tell me, then why was it included it in the software? What you seem to be telling me is to perform step 4-6, doing a First Pass (generating the first pass stats file), opening it, then checking the compressability, perhaps changing the resolution after that, depending on the result? Basically what I understand from what you write is that I should ignore the Aspect Ratio Error box, unless it goes way off?

I am sorry, but this is pretty confusing to me. Perhaps I have been unclear about why I ask, what I am trying to achieve?
All I want is to create a DivX that as closely as possible resembes the DVD. Whether I output the DVD or the DivX from my graphics card's out to my TV, or watch it on my monitor (which I never would set to anything but a true 4:3 relationship) I want the movies to look as alike as they can, the same amount of black borders, etcetera.

I want as little of the original image cropped off as possible. If the DVD doesn't have perfect Aspect Ratio, which I of course can't be sure of if I didn't shoot it myself like Wef pointed out, then fine. I just want my DivX to look like the DVD. I have paid attention the to Aspect Ratio Error box and have clicked Follow the ITU-R B.T. 601 Standard. Is this so bad of me?

Dark-Cracker
11th May 2002, 19:30
can someone could say me how to code a smartcrop all fonction (getting the new crop size for getting an aspect error = 0.0%) ?

thank u .

TheWEF
11th May 2002, 22:05
Originally posted by Swan
I just want my DivX to look like the DVD. I have paid attention the to Aspect Ratio Error box and have clicked Follow the ITU-R B.T. 601 Standard. Is this so bad of me?

to (hopefully) finally clear things up: :p

for quality (~filesize) reasons you should use mod 32/16 or 16/16.
ok. - so...
in many cases this might introduce an aspect error.

it is hard to objectively measure this error.
the error percentage in gknot is not an absolute value, it's calculation is based on assumptions. e.g.: "the value is correct if the dvd has been mastered following ITU standard, in gknot ITU is checked, (i watch my movie on a perfectly adjusted computer monitor in a desktop resolution with 100% square pixels and i adjust my tv-out so that the usual overscan effect is compensated)."
if all this is true and the error-percentage is zero your avi will have exactly the same AR like the original film material. in this example this is not the same AR that you see when watching this dvd on the same monitor with a software dvd player (powerdvd, windvd,...) - in this example your avi is correct and the software dvd player is not.

so you have this ar-error value. and with your experience and "your own judgement" you have to interpret it, maybe alter it (check or uncheck ITU = changing your assumptions, checking the range...).

now you have the choice:

a) live with it, but keep the complete frame.

b) crop a little more (smartcrop) to eliminate the ar-error.

c) do something in between, manually.

but never change your (horizontal) resolution just to get rid of the ar-error. it's not worth it! balancing resolution, filters, bitrate and compressibility is what video encoding is all about! that's the art!
AR is a minor problem that you should not put too much brain into.

IMHO if your AR is unnoticeable wrong - it is right.


so. i swear this is my last post in this thread! :devil:

wef.