low I-frames? also - about anamorphic

[Blue_MiSfit]

Hey all,

I have been doing a lot of encodes with the new dev-api4 tree, and I have been getting for the most part excelent results, but I cant help but think they could be better. My stats show my I-frames as being very low, IE 1361 I frames out of 111,806 total, with about 3x as many B frames as P frames otherwise.

I'm doing 1.2 gig rips with (roughly) 128k vorbis audio, so that usually averages out to about 1.1 gigs for the video.

settings-wise, I always use MPEG quant, highest quality VHQ and motion prediction, GMC, Qpel, adaptive quant, and a bvops setting usually around 3 / 1.5 / 1.0 . Additionally, I always enable Chroma motion and usually set my Maximum I frame interval to 200. On another note, I usually set BVOP sensitivity for the actual movie to 2, and to 5 for the credits (along with quant @ 31 etc...)

I try to keep the bits/(pixel*frame) ratio in GKnot in the mid to low .20's, so I usually end up running somewhere in the high 600s for my horizontal resolution, depending on the length of the movie. The quality is good, but I really think it could be better. So do I have too few I-frames?

On another slightly unrelated note, I have been trying to figure out how to get Anamorphic mode to work. Now, before I get yelled at for not searching, I have searched, but haven't found anything that has specifically answered my questions. What do I specify in the resolution tab of GKnot if I want to do an anamorphic encode? Do I leave the input pixel aspect ratio at 16:9, or change it to 1:1? Finally, what do I select in the Aspect Ratio tab of XviD? Do I change pixel aspect ratio to NTSC16x9 or do I use picture aspect ratio at 16x9? confusing!

I did a test encode with Mad Max using 1:1 in GKnot, and 16x9 pixel in XviD, but the image definatley was not anamorphic.

Sorry for the lengthy post, but I'm freakin confused all of a sudden :)
~misfit

[sysKin]

I don't get it - you want *more* i-frames? Why? The more i-frames the lower your quality. You only ever need them for seeking in the file, and if you don't want to seek, you don't want them at all...

[Blue_MiSfit]

really?

I did not know that :)

Thanks though, and any info re: anamorphic?

~misfit

[GolovachLena]

Originally posted by Blue_MiSfit
I always use MPEG quant, highest quality VHQ and motion prediction, GMC, Qpel, adaptive quant, and a bvops setting usually around 3 / 1.5 / 1.0 .

If you're so concerned about quality, you could check adaptive quant off and lower bvops settings, 1/1.50/1 for example. I think adaptive quant, trellis is big evil if you're making 2 cd rips. And even do not use gmc and qpel which could lead to artefacts and incorrect playing under some circumstances.

[Teegedeck]

Originally posted by GolovachLena
I think adaptive quant, trellis is big evil if you're making 2 cd rips.
No, it isn't.
Originally posted by GolovachLena
And even do not use gmc and qpel which could lead to artefacts and incorrect playing under some circumstances.
No, it doesn't. How did you come to such a conclusion?

@Blue_MiSfit: Using three B-frames is not always safe, thus the default is=2.

Switching off trellis and AQ will do more harm than good in the end; also GMC and quarterpel are not vital, but especially quarterpel makes for a big improvement in subjective quality. Keep them in.

Instead of switching off trellis and AQ, use a different custom-matrix like the HVS-best-picture or Andreas 78er. The result will look much better. BTW, search the forum for sysKin's posts if you want quick, reliable advice on a topic.

---
Anamorphic encoding: It has been said several times that XviD's own PAR can only be displayed correctly if you use mplayer or NeroDigital's DSF for playback.

Matroska's AR-feature is displayed correctly if you use ffdshow with the 'directshow overlay' option activated. Use mkvmerge to set the AR for matroska.

(Edited: mix-up in usernames)

[Blue_MiSfit]

Anamorphic encoding: It has been said several times that XviD's own PAR can only be displayed correctly if you use mplayer or NeroDigital's DSF for playback.

okay, so I understand how to playback anamorphic encoded content, but my question still remains. What do I set the input PAR in gknot to, and what do i tell XviD in the AR dialogue? (sorry about the non specific terms, I'm at work and dont have xvid installed here)

[mikeX]

Matroska's AR-feature is displayed correctly if you use ffdshow with the 'directshow overlay' option activated. Use mkvmerge to set the AR for matroska.

the latest XviD decoder supports it as well

Concerning the par settings:
there is no standard answer, the setting could be different each time

:edited out: sorry not what you asked...:edited out:
you can do as you said through gordian knot, set 'Input PAR' to 1:1 (don't change 'input resolution') and then set xvid picture AR to 16:9 or whatever, but that's not gonna give that much horizontal difference

if you want to downsample to a custom anamorhically compressed resolution (e.g. 544x432) you'd better do it manually in your avs script (e.g. LanczosResize(544,432)
::edit: if you still wanna do it from Gknot you'll have have to change 'Display AR' approprietly...
then set your xvid PAR with a simple calculation:
(assuming you have a 16:9 source)
432 x 16 / 9 = 768
768 / 32 = 24
544 / 32 = 17 (since they are both mod 32, if one is lower (e.g. 8) you will have to divide both with that number)
voila, you have your custom PAR: 24/17 (setting the xvid picture AR @ 16:9 should do the same, but i haven't really tried it)

the setting is really usefull in the case where you don't want to downsample your image, where the original horizontal resolution is 720 instead of (for a PAL 16:9 movie) ~1024. then you can set xvid's PAR 16:9 PAL (same thing as above for picture AR)

(ps: thanks for the info on overclocking...)

[EthanoliX]

From my point of view it is better to leave BVOP sens. at 0 if your working at high bitrates. Maybe lowering it might increase quality.

Even a BVOP sens. of 5 results in a huge increase of b-frames. And my impression ist that this might hurt quality, sometimes even for a 1CD rip.

[Blue_MiSfit]

I only have the sensitivity set to 5 for the credits, since i really dont care about them. For the rest of the movie it was set to 2 originally, but I have it back to 1 now, which combined with hvs-best seemed to really help quality. My Road Warrior rip is exquisite.

Thanks all
~misfit

[Teegedeck]

Originally posted by Teegedeck
Anamorphic encoding: It has been said several times that XviD's own PAR can only be displayed correctly if you use mplayer or NeroDigital's DSF for playback.
Originally posted by mikeX
the latest XviD decoder supports it as well
At least my RC1-DSF doesn't. I don't see that in the changelog, either.

[Koepi]

Tee:

if you use avi as conatiner for AR files, xvid dsf copes correctly with it. Same for mkv.

Regards
Koepi

[Teegedeck]

Now I've found the entry in the changelog from 2004-01-17 01:09:01 that states AR-information is forwarded.

But yesterday I tested it with an AVI-container and neither entering the info as DAR nor PAR seemed to make a difference for the DSF during playback.

I'm going to give it another go when I get home.

Edit: A-ha; do you use that xvid.ax sysKin linked to in the 'flipped video'-thread? I ignored that one till now. And here I go telling people to follow anything sysKin posts...

[mikeX]

@ Teegedeck

I was actually reffering to the matroska AR feature, I didn't know there was support for Xvid's own PAR/AR until koepi just mentioned it :) GREAT!

I also tried out the AR feature (instead of PAR). It worked great (and saved me the trouble of calculating PARs - don't know why i never used it in the first place)

[sysKin]

Originally posted by Teegedeck
But yesterday I tested it with an AVI-container and neither entering the info as DAR nor PAR seemed to make a difference for the DSF during playback.

I'm going to give it another go when I get home.I explaied that somewhere but I can't find it anymore. Anyway, the answer is that decoder reads AR information written to AVI header but encoder does not write them. This is because I didn't want any people to be any more confused - these information are not copied when stream is muxed anywhere and might dissapear in other circumstances too.

It's easy to add it... but I plan to do that after 1.0 final is out.

[vijv]

i did an encode with 3 bframes and a sensitivity of 10. the result was not very crisp and seemed a litter softer. i think you should stick with the defaults unless you are doing very low bitrate encodes.

I also did an encode with HVS_best_picture and it seemed very strange that P frames got quants of 2, whereas I frames got quants of 3. is that normal?

[vass-iliskus]

Originally posted by sysKin
I explaied that somewhere but I can't find it anymore. Anyway, the answer is that decoder reads AR information written to AVI header but encoder does not write them. This is because I didn't want any people to be any more confused - these information are not copied when stream is muxed anywhere and might dissapear in other circumstances too.

It's easy to add it... but I plan to do that after 1.0 final is out.

Hi sysKin,
if I get it right from your words above and your conversation with shitowax here (http://forum.doom9.org/showthread.php?s=&threadid=68117&perpage=20&pagenumber=3), the only way currently avaiable to make XviD DSF use custom PAR is to patch biXPelsPerMeter and biYPelsPerMeter in VIDEOINFOHEADER/BITMAPINFOHEADER inside AVI file ? It works indeed (and the encoding looks marvellous, thanks to you and the others), I just wonder whether there is another, more "official" way, and do you have plans to change the way it all works ?

P.S. The question is about AVI only...