Overlay bug: XviD and jumping dots

[Lastique]

Using BSPlayer for playing an XviD-coded movie I've noticed red and blue square dots jumping over the movie screen, mostly in the
dark places of the picture. Playing the movie in other players (Windows Media Player for example) or disabling Overlay in BSPlayer
solves the problem, but in the last case CPU usage raises. Overlay Mode is Mode 1. Mode 2 doesn't work, when I try to select it
BSPlayer crashes with the following messages:

BSPlayer v0.85.492, Unhandled exception at EIP: 004923A4
If you click 'Close' application will be terminated.
Please report this info to the author with description what were you doing.
Access violation at address 004923A4 in module 'BPLAY.EXE'. Read of address 00000000
EAccessViolation
, 0x004B362F, 0x004B38A9, 0x004520BF, 0x0044972A, 0x77D43A68, 0x77D43B37, 0x77D43D91, 0x77D4438C, 0x004C48B7, 0x77E814C7

Bytes at EIP: 8B 00 FF 50 60 33 C0 5A 59 59 64 89 10 68 CB 23 49 00

By the way, what's the difference between these two modes?

PS: Tried many versions of XviD (up to alpha 14.12.2002), tried ffdshow (up to alpha 13.12.2002) - the result remains the same. Detonator 41.09.

[Lastique]

Some addition. As I figured out the problem is with overlay in YV12 only (other apps showed the very same dots only in this colorspace). If overlay is in YUY2 no dots are shown. So WMP seems to be not using overlay at all or converting the colorspace on the fly. And I think DivX decoder shows in YUY2, so no bots too. Now that's the question: is there any way to force bsplayer to overlay only in YUY2?

Oh, by the way. I use Det. 42.01.

[bst]

Not yet it will be in next build. But I think that latest vesion of Xvid decoder works ok in YV12 also it defaults to YUY2.

[trodas]

Actually, bst, please, what are the diferences between mode1 and mode2 at all? Mode1 working with latest build and the build before it for me too, but mode2 keep crashing.
Is there any quality of speed resons, why use mode2?
Whats the diferences whatsover???

Results from trying mode2 on latest build with Detonators 41.09 and Win2000 SP3 + hotfixes:

BSPlayer v0.86.493, Unhandled exception at EIP: 356B3654
If you click 'Close' application will be terminated.
Please report this info to the author with description what were you doing.
Invalid floating point operation
EInvalidOp
, 0x004B48EF, 0x004B4B69, 0x004520BF, 0x0044972A, 0x77E3A2B8, 0x77E1

Bytes at EIP: DE F9 8D 45 F0 DD 1C 24 50 E8 EC CB FF FF DD D8 83 EC

[-Yabba-]

Hello all ... I have a similar problem with Xvid, my graphics card ( ATI all-in Wonder 128 ) won't display overlay in YV12. It only shows green lines. Xvid movies always default ( for me at least ) to YV12 and I can't find a way to make them use YUY2. Overlay Mode 2 makes BSPlayer crash, and finally disabling overlay makes the movie watchable, but really bad ... it looks pixelated. Besides, WMP 8 shows the movie perfectly ... but I can't have subtitles and anyway I prefer BSPlayer and not Windows-crappy-Media-Player.

Am I doing anything wrong with BSPlayer ? Or maybe I'll have to wait for new build to force it to use YUY2 ?

BTW, I tried some different Xvid implementations, and also FFDSHOW, with same behaviour. FFDSHOW also uses YV12, even in Divx content ... when Divx codec used YUY2 normally.

Thank on advance for your help ...

[KennyMcFr]

I've just updated my ffdshow to the latest ffdshow-alpha, and I think I've found a way to force YUY2 with it (or at least to prevent it from using YV12).

Open ffdshow options and select "Codecs" menu.
In "Supported output colorspaces" settings, just uncheck"YV12".

The video should then play nicely in BSplayer's overlay mode 1.
I also used to have jumping dots while using ffdshow (I use these filters to play most of my videos: DivX3, 4, 5 and XviD), and looks like they've just vanished! ^_^

[bluedan]

Seems to be the only solution so far.
People over @ doom9 forum reported same error with same solution a month ago.
Question is -and has been over there, too- if your graphic card reports to support YV12 overlay but fails to do so.

It's irritating to me, why YUY2 overlay should outperform(=no display bugs) YV12 because the latter is much faster. It uses 12bit per pixel instead of 16bit used by YUY2, though at a lower bandwidth.
Actually, if a graphic card manages YUY2 overlay so why the heck it doesn't do YV12?
Also most DVD/MPEG2 data is stored in YUY2 mode defaulting in YUY2 overlay display -who wonders why- no color space convertion needed.

If you have an MPEG4 stream such as XVid or DivX, it will be decoded in YV12 (YUV 4: 2: 0) internally, because it's restricted in MPEG4 specs, and then converted to YUY2 if requested as output. Working in this sequence why does that error described above not occur in YUY2 also, for it's the 2nd step in the row???

Look here: http://forum.doom9.org/showthread.ph...DivX+YUY2+YV12
and here: http://forum.doom9.org/showthread.ph...er+output+yuy2.

I'm not 100% sure about current default Xvid decoder output, BST.
I think it still defaults to YV12 in the first line. Frankly, I was unable to collect further information about current build. :roll:

As an alternative you may use DivXg400 filter to force YUY2!

Lately, as until now BST did not reveal the nature of overlay 1 and 2, I assumed that mode 1 was YV12 and mode 2 YUY2...

[bluedan]

Originally Posted by bluedan

I'm not 100% sure about current default Xvid decoder output, BST.
I think it still defaults to YV12 in the first line. Frankly, I was unable to collect further information about current build. :roll:

Lately, as until now BST did not reveal the nature of overlay 1 and 2, I assumed that mode 1 was YV12 and mode 2 YUY2...

Well....now after checking via ffdshow info tab, XVid output always seems to be YUY2, no matter if overlay mode 1 or 2 in bsplayer nor decoding done by ffdshow=libavcodec.dll (decoding done by XVid unchecked) with YV12 output unchecked.
Maybe ffdshow does not show correct info, but I don't think so...
Further investigation shows my NVIDIA TNT2pro to be able to display in 9(!) different colorspaces, amongst them YUY2 and YV12, at least according to Sisoft Sandra DirectX information.

It's DirectX 8.1 and XVid dev-build 15-01-2003(koepi) and I don't understand nothing now evermore :shock:

[KennyMcFr]

I don't know much about these YV12 and YUY2 things, and I just discovered this today in ffdshow options after reading this topic.
Before that I also found another way to avoid the jumping dots problem using BSPlayer's overlay options (note: I use BS 0.86 b493).

Dunno if this solution can interest anybody or if it has already been given, but well... Please just bear with a beginner like me! ^_^

So, to avoid the problem you can also enable both the overlay in mode 1, and RGB overlay in RGB32 mode (force RGB option disabled).
You will then get a video which plays nicely... but upside-down!
So you'll still have to find a way to flip it; however the jumping dots problem disappears.

[bluedan]

If you enable RGB overlay then you have to have a powerfull CPU around 1GHz or more, for this colorspace conversion takes a lot computing.
Flipping video image is no problem, ffdshow has many options.
Open Miscellaneous configuration tab, tick Flip video box.

[trodas]

Very insteresting discussion, i added a comment on the Doom9 forum one, tought it still fade to gray my basic question - what the hell is a diference between mode 1 and mode2...

...altrought it seems to be obivous, that mode1 forcing YUV 420 (what is fastest and the right format for mpeg1 and mpeg4 formats (DivX, Xvid) -and maybe and probably also for other WMF formats, since IIRC only mpeg2 will use YUV 422 (YUY2), right? :?:


bluedan - one thing is, what driver tells you, and another what is realy supported. For example, on every low resolutions, including and bellow 512x384 video get verticaly halved :evil:
So, it can very good suxx when come to it.
But i believe that YUV422 works well on my GF2Ti, since DVDs get played into this - so, altrought there is no point (only slow-down) to forcing BSplay to use this format, it SHOULD work.

And maybe it should be also more properly marked, even as YUY2 and YV12 is better that Mode 1 and Mode 2, what tells you s*it about it
Tought im familiar of naming the colorspace to its quantization of signals, so, YUV 422 and YUV 420 fit best, IMHO 8)