Some bugs in xvid codecs

[libredr]

Hi !
After using DivX3.1;), then DivX4, I decided to use the xvid codec, both for quality and philosophical/political reasons (it is GPLd !)
I had some success, but some builds I found on Internet had some problems. First, I wish to thank people who compile the codec, and make them available to everyone.
Here are my problems :
- the pink problem has already been described, with builds after the 30/01 (I have an Athlon TB 1.4) People who want to reproduce it should encode the logo from columbia (for example, at the beginning of final fantasy) : the image showing a girl with a torch in the hand has pink areas until the word "columbia" appears.
- the Koepi's version of 9/2/2 has a strange problem : the image is aliased, and never smoothed by the player, contrary to other versions.
- the uManiac version of 10/2/2 does not correctly do the first pass on my system : instead of using the highest quality, the 1st pass image is very blocky, as if it was created with a very low bitrate. I guess this is related to the important changes in the codec's setting tabs ? However, it seems that the pink bug has disappeared (but it is difficult to say because the image is very blocky).
Tonight I will try the 11/2 version from Nic.

For prople looking for binary codecs, here are the three links I have found useful :
uManiac http://www.heimsnet.is/kg/
Koepi http://www.roeder.goe.net/~koepi/
Nic http://xvid.stormpages.com/ ou http://www.freewebz.com/xvid/
some useful information can be found on http://www.geocities.com/avihpit/xvid/cpu-opt.html

Once again, thank you guys !

I hope that Gordian Knot will support xvid one day, that would be brilliant !

Another question : are there people here that use xvid under linux ? Do you use transcode or ffmpeg ? Is there any step-by-step guide about how to use them ?

Thanks for your replies.

[-h]

Re: blockiness - are you using the version with a 2-pass tab? If you specified a frame to begin the credits (i.e. it's not set to 0), every frame after the credits frame will be encoded with the "quantizer for credits" value, in the 1st as well as 2nd pass.

I'll make this simpler..

I hadn't tried XviD for a week or so, but I haven't seen any pink/green blocks with the current build (guess I missed all the fun!)

-h

[grug2k]

Theres a bug relating to the XVid decoder and WinXP's explorer.

The windowpane to the left hand side of explorer windows makes previews of AVI files. Should you happen to click on an xvid AVI (using the xvid fourcc), explorer tries to create a preview but can't. It leaves the file in use and also explorer starts using up 95% or more CPU time.

Its quite annoying to have to end task explorer several times. :)

I only mention this because it used to happen on DivX, but 4.12 doesn't seem to have the bug, so I presume theres a way to fix it and its not an M$ problem.

[Franko30]

to -h:

I noticed the blockyness in Nic's Feb11th binaries, too.

So I recorded a one minute test clip and made three 2pass test encodings as follows:

All used H.263/Motion Search 5/ 2pass, 2nd pass int. with desired filesize 6651.

1st Test: Credits start 0/Quant 16. Resulting filesize 1382, absolutely blocky...

2nd Test: Credits start 0/Quant 2. 1st pass OK, 2nd pass delivers Error message "Credits size larger than filesize".

3rd test: Credits start 1000/Quant 16. Both passes OK, good picture Quality before frame 1000, blocky afterwards. Resulting filesize 6698.

Conclusion:

Credits start 0 doesn't "turn off" Quant 16, but uses it for the whole encoding, thus the blockyness...

So you have to specify an end credits start frame (for example the last 5 frames), even if you don't use the end-credits feature (old movies without credits at the end).

Frank

[Koepi]

I found & solved the bug on 1st pass you described, watch on the thread "My binaries" of nic.
The aliasing stuff directly comes form the XviD core - I didn't change anything on that so it should appear on all builds after 09.02.2002...

Regards,
Koepi

[libredr]

Thanks everyone for your replies.
Yes I used the same .avs file than I used for DivX4 to test xvid, and I had the end credits turned on. I will try without this feature.
However, I am positive there was still the pink problem with the 5/2, 7/2 and 9/2 builds (on my system). These are not really plain blocks, but areas with a dominant "more" pink colour (sorry for this example of crappy english !) The colorimetry was then ok (the transition was very obvious) when the Columbia word appeared. I can send some screenshots when I return home. I suppose this is the same "pink" bug than previously described.

For Linux users, I can answer to myself after doing some searches :
There is : http://www.doom9.org/divx_linux_guide.html using transcode and perl scripts
and a nice GUI doing ripping and encoding : http://www.exit1.org/dvdrip/

[kastro68]

I have done some short test clips myself with Xvid 9/2/2002 using an average bitrate of 800

I found the codec to produce extremely unacceptable blockiness for this short clip. Not only was blockiness existent, but so was deringing around moving objects. Luma Masking was not turned on.

I have previously done a whole movie with 7/2/2002 which appeared more acceptable apart from the pink.

Is this problem caused by the fact that i used a short clip or is it because of a bug in 9/2/2002.

Oh, and if a dvd has a huge scratch how can i skip past the defect.

It occurs at 400 megs on the 4th Vob file of the FOTR dvd. I tried using Vstrip but it seems it cannot skip past the error...the dvd works in a dvd player though.

Sorry about posting this question in the wrong place.

Cheers

[libredr]

Hi everyone !

I have just tested the new 11/2-3 release from Koepi's site and it works very well : the pink bug is completely removed. However, I have still the aliasing bug. The image is not smoothed by the player (like WMP or PowerDivX), contrary to the january version of the codec.
But, since both images are aliased in Virtualdub, I conclude that the smoothing is not encoded, but probably a kind of command which is given by the avi file, command which should be written by the codec but is broken in latest releases, or maybe there is something in this codec which prevents them to do the anti-aliasing. Am I right ?

[Acaila]

Are you sure you're not using XviD's DSF? DivX4's DSF has post-processing, XviD's doesn't. By default the first is used, unless you manually changed it...

[libredr]

No, I just installed xvid version of the codec only, like any others (did not install xvid.ax) And in the same conditions, the 30/1 version is smoothed by the player, not the today's version. Is there something wrong ?

[Acaila]

What's the FOURCC code of the movie you're testing with?

If it's "DIVX" then smoothing should occur and you've got a problem.
If it's "XVID" then there isn't supposed to be smoothing, and changing it to "DIVX" will give you back the smoothing.

You can check this with Doom9's FOURCC changer which you can find in the downloads section.

[libredr]

It is that exactly, I used xvid FourCC. Thank you Acaila.
Otherwise this codec is brilliant. I think I will use it as daily basis now, very soon. I want to try it on my Linux system now.

[libredr]

Allright, now I have understood what the FourCC was (thanks to http://www.webartz.com/fourcc/). So the player does not find a known FourCC (because xvid is not yet officially recognised) and does not perform the anti-aliasing.
One remark : I read that Microsoft is responsible for the registration of FourCC. Will they give an official FourCC code for xvid, which is GPL (their so-called "cancerous" enemy) ?

[western shinma]

That's not quite right, the player does find a known fourcc since xvid is installed. The xvid decoder simply does not have any postprocessing yet, but it should be implemented eventually.

[Ripe73]

HI!

I cant save the first pass statsfile where i want it and its easy to overwrite it if you do the mistake and not change to"2 pass second pass":D ,i did.
When i use the Debugview the vidoeframes is not in synch with Virualdub's videoframe(Debugview +500frames)i dont know if is a bug.

Hope you guys can fix it

Thanks for a good job

XviD-12022002-1