Possible Xvid bugs for Koepi and -h

[kastro68]

Guten Morgen,

Well, i think there may be another diagonal bug with both Koepi's 17/2/2002 and the Nic's 18/2/2002 build...i haven't checked to see if you have put a later build up just yet.

Recently i have had a relapse of the diagonal bug which shows itself when the resolution is not a multiple of 16. I was pretty sure this problem was solved once before, since i have previously done an encode using a resolution of 656x270 with xvid and had no problems.(using koepi's 17th build)

The bug appeared when i used a resolution of 644x286...when i did 2 passes there was colour but a multicoloured diagonal(not purely green like last time) from bottom left to top right. Then after that i kept changing the resoln and did a cbr encode to test it, and it came out black and white with the diag problem. I found only resoln with a multiple of 16 worked. (problem appeared with both koepi's 17th and Nic's 18th)

Could it this be because the dvd was in NTSC format? I didn't have any problems with Pal dvds just yet. I used force film in dvd2avi, and i also used smart deinterlace when i saved the avs file.

I don't think i have isolated the problem yet, and don't really know what is going wrong.

Could be because i just keep installing a later build without uninstalling the old one? Unlikely i'd say.

Oh, and with Koepi's 17/2/2002 build there is something wrong with the opening and closing credits feature. It works but very badly.

I'll check now to see if a later build has been released,

ja ne

[gldblade]

>Oh, and with Koepi's 17/2/2002 build there is something wrong with the opening and closing credits feature. It works but very badly.

What do you mean badly? Bad quality? If so, maybe your settings are a little low. For example, using the default of 30% gives me really ugly looking results.

[kastro68]

I am certain there is a diagonal bug with 21/2/2002

I just tried to reincode a vcd file into avi, but i chopped off the edges a bit where there was some black and used a resolution of 350x236 which is NOT a multiple of 16.

Is this a bug with the codec? Or is it that this codec was not designed to be used for resolutions that are not a multiple of 16? It would be very disappointing if this codec only works for multiple of 16s.


Regarding the credits...i get trails or leftovers from the previous frame showing up, this occurs at the opening credit. The credits appear very blocky too, but this might just be that i am not allocating enough bitrate to the credits.


Ja ne

[kastro68]

Just found another bug.

I didn't mess with any of the quantizer settings. I tried to encode a 200 meg vcd file to a 140 meg xvid file...using 2 passes of course.

The file came out with a size of about 184 megs instead of about 140 [120megs vid, 20megs for audio]. There are 25 eps to this series and i am trying to fit 5 eps per 700 meg cd.

Episode 1 came out to be 138 megs...which is alright
Episode 3 came out to be 139 megs...which is alright

BUT, Episode 2 came out to be 184 megs...which is unusual.

I even encoded it again and it still came out about 188 megs also.

Is this a bug?

[-h]

Hi kastro68,

Could you post an extract of the dbgview output for the oversized encode? It sounds like a bug, but I'm not sure where it could be coming from..

-h

[kastro68]

The bug does not occur with the 21/2/2002 build when i use h.263 quantization.

The oversized file only occurs with Koepi's 17/02/2002 when i use modulated quantization.

BUT, the 21/2/2002 build bleeds green and pink ooze from the right edge where i cut the video to get rid of a black edge. However i think this is a result of the 4.12 DSF.


Oh, and i just recently got a resolution of 348x236 to work...which obviously isn't a multiple of 16. I just don't understand why other resolutions may not work. It is irritating to have to do a test encode to see if the resolution is feasible every time.

Does anyone know if this is a bug or not?

[-h]

The oversized file only occurs with Koepi's 17/02/2002 when i use modulated quantization.

Darn, the overflow fix I merged would only really fix undersizing problems. Think I'll let Koepi play with this one :)

BUT, the 21/2/2002 build bleeds green and pink ooze from the right edge where i cut the video to get rid of a black edge. However i think this is a result of the 4.12 DSF.

Oh dear. I'll ask the core guys, but internally the codec should be pretty content down to any multiple of 2 pixels. Unless our padding/mv restriction code is stuffed.

-h