ffdshow tryouts project: Discussion & Development - Page 338

haruhiko_yamagata · 23rd February 2009, 15:40

Quote:

Originally Posted by albain

Yes this is my fault (this was not an accusation by the way, I suspected this was an old bug because it would not crash under debug mode. Those bugs are the most difficult do trace)
Thanks I got your email. How did you reproduce without debug mode ?

Following your guidelines, I think this is fixed by now : revision 2716

I could reproduce the crash with debug build, however, if CoreAVC is used in the filter graph, we can't use debugger.
I used Debugview for Windows and many DPRINTF. It's the classic.
Thank you for the fix.

haruhiko_yamagata · 23rd February 2009, 15:44

Quote:

Originally Posted by clsid

Does that mean that the dvd menu positioning issue is caused by ffdshow sending bad interlacing info to the next filter in the graph? But since ffdshow gets the same bad info from the MPEG-2 decoder, it is not really ffdshow's fault?

The aspect ratio rather than interlacing flag. The position is calculated in the application (MPC-HC in this case), ffdshow has to send some info to video renderer or somewhere so the the application can calculate correctly.

Quote:

When ffdshow is only doing the processing, it is not able to (use the sequence flags to) detect whether it currently is processing a 1 frame menu, right?

It's impossible as the sequence flags are obtained by ffdshow custom code in libmpeg2 and libavcodec.

madshi · 23rd February 2009, 15:47

Quote:

Originally Posted by haruhiko_yamagata

It's impossible as the sequence flags are obtained by ffdshow custom code in libmpeg2 and libavcodec.

I think you could go down (you would say "up"

) the filter chain by browsing through the connected pins to find the MPEG2 output pin and then ask its connection media type and grab the codec private data from there, which usually contains the sequence headers. Isn't that what you need? A bit ugly, but it might work...

mark0077 · 23rd February 2009, 19:05

Quote:

Originally Posted by haruhiko_yamagata

As for the positioning of DVD menu, if I uncheck "Set pixel aspect ratio in output media type" the problem is fixed. The AR is still correct. Tested with overlay mixer, VMR7, VMR9.

Damn, unfortunately setting this option to any of its 3 available options has no change on my menu highlights. I have also tried playing with settings like the mpc-hc "Keep aspect ratio" setting, and the ffdshow resizer on and off, ... the highlights seem a few pixels off, or slightly garbled in some cases.

Tis a pity. In my opinion, if ffdshow relies on internal data like you say, flags set by the internal decoders, and gains an advantage by doing this, I think every effort should be made to try and get that information from raw video also. I will leave the implementation upto you guys but it just sounds like the right thing to do... ie not favour internal / external decoders if the information can be gained using other methods. (although in this case, decoding WITH ffdshow seems to be causing the ugly highlights on 1 frame and moving menus, decoding WITHOUT ffdshow causes the freeze/crash in 1 frame menus, and decoding WITHOUT ffdshow in one frame screens with de-interlacing incorrectly enabled causes freezes / crashes).

tetsuo55 · 23rd February 2009, 20:01

Shouldn't the relevant information be passed on between each step in the filter graph?

Information like:
-AR(in all its flavours)
-Framerate
-Interlace flag

Maybe even more information is usefull, if any value is no longer true(in the case of deinterlacing) the part of the chain that caused the change in the value should update the information to the new status and pass that along.

haruhiko_yamagata · 23rd February 2009, 23:43

Quote:

Originally Posted by madshi

I think you could go down (you would say "up"

) the filter chain by browsing through the connected pins to find the MPEG2 output pin and then ask its connection media type and grab the codec private data from there, which usually contains the sequence headers. Isn't that what you need? A bit ugly, but it might work...

SEQUENCE_START/SEQUENCE_END flags indicate that the frame is a start/end of a sequence. It is not possible to use media types, because media types are not attached to all frames.

lych_necross · 24th February 2009, 09:55

Quote:

Originally Posted by leeperry

kado gave a link in the previous pages where it's pretty clear that mp3 being lossy, you wanna decode it in the highest resolution to avoid roundings....that's what I've been doing for a long while. plus I don't use KMixer

I re-read that link and I see what your saying. I normally don't use gabset for mp3 playback (I usually use foobar)

haruhiko_yamagata · 24th February 2009, 13:18

Quote:

Originally Posted by mark0077

Damn, unfortunately setting this option to any of its 3 available options has no change on my menu highlights. I have also tried playing with settings like the mpc-hc "Keep aspect ratio" setting, and the ffdshow resizer on and off, ... the highlights seem a few pixels off, or slightly garbled in some cases.

A few pixels? Then it's quite usable.
With some special DVD (wrong encoding IMO), the menu is off by 10% of the screen height. I found I can avoid this by using DVD navigator's aspect ratio instead of decoded aspect ratio.

mark0077 · 24th February 2009, 13:27

Quote:

Originally Posted by haruhiko_yamagata

A few pixels? Then it's quite usable.
With some special DVD (wrong encoding IMO), the menu is off by 10% of the screen height. I found I can avoid this by using DVD navigator's aspect ratio instead of decoded aspect ratio.

Oh definitely usable but consistently wrong for every DVD I have... Its not just a few pixels off really you can see in the screen shots where the perfect rectangle that mpc-hc draws isn't drawn nearly as good when using ffdshow decoders. The corners get blurred and the lines are skewed and a few pixels off. MPC-HC seems to get it perfect each time with every DVD I have... Not a big issue, at least better than the crashing when not using the non ffdshow decoders so for the moment I will put up with ugly highlights until someone would like to take a look into the code

Actually I do notice, on this Queen DVD menu where you can select the audio output, PCM Stereo, DTS, or DD I think it is. The three options go from top to bottom so

---------- top of screen -------------

Option 1 - Highlight 1 should be here
< Position of option 1 highlight (ie. skewed toward centre)

Option 2 < Position of option 2highlight (90% perfect)

< Position of option 3 highlight (ie. skewed toware centre)
Option 3 - Highlight 3 should be here

---------- bottom of screen -------------

When I look at where the rectangles around these options are, I notice they are all skewed / offset towards the centre of the image (horizontally). Those further from the center get offset more so the further from the horizontal centre, the worse any menus will look in this combination.

So i think you are right in saying the aspect ratio flag has something to do with this problem. Its like the highlights are being displayed on an overlayed image but in slightly different aspect ratio window to the underlying image, like as if the overlay window isn't being given the right AR. I don't know how subtitles / highlights work but this sounds like a reasonable explanation.

Do you see any easy solution.... what part of the ffdshow AR code might be at fault? I assume an AR is also supposed to be output with these subtitles / highlights, or at least taken into account when rendering them. Seems as if it is not in fact being taken into account correctly. Not a big deal but still would be nice to see it fixed now that we have an idea what it could be.

EDIT: Unfortunately build 2718 doesn't fix the above for me. Thought it was going to looking at the change list :P I thought fiddling with ffdshow resize settings, and turning it on or off, or editing the ar output option might have an effect but not for me. All highlights are stretched towards the centre of the screen by a few pixels, as I said, the further the highlight should be from the centre, the more it seems to get offset.

haruhiko_yamagata · 24th February 2009, 13:30

Quote:

Originally Posted by iron2000

I don't mean using the DScaler deinterlacing filters in ffdshow.
I mean using ffdshow as a filter in DScaler, to postprocess the picture in DScaler.

Was able to do it in some version of ffdshow way back.

OK, I'll take a look later.

flanger216 · 24th February 2009, 16:24

This has been bugging me for a while:

If you use libavcodec to decode MP3s, and you have 16, 24, 32 and 32-bit floating-point enabled as outputs, then ffdshow outputs 16-bit. If you deselect 16, 24 and 32-bit integer as output formats, then libavcodec outputs to 32-bit floating-point. So in the latter case, is libavcodec truly decoding to 32-bit floating-point, or is it still decoding to 16-bit and then merely padding the output to 32-bit floating point?

This also applies to AC3, DTS, Vorbis and all the other lossy (bitdepth-less) codecs that libavcodec can handle.

I only ask because libac3, libdts and libmad all seem to output to the highest bit-depth available, whereas libavcodec seems to prefer using the lowest available.

leeperry · 24th February 2009, 20:09

Quote:

Originally Posted by lych_necross

I re-read that link and I see what your saying. I normally don't use gabset for mp3 playback (I usually use foobar)

the mp3 decoder in foobar is very unforgiving IMHO, the sound is just flat flat flat...I'll take libmad(Gabest or ffdshow) anyday over foobar for mp3.
plus you can't process VST/winamp plugins in 32 float, it's 16 integer or nothing....which also sounds ugly

Quote:

Originally Posted by mark0077

leeperry: no I have never been able to get menus working with HR. The three bugs I talk about in my post above are all related to ffdshow and can be fixed within it I am sure. The crash I see is, as haruhiko_yamagata says due to the way ffdshow with yadif works, and only seen in dvd menus where single frames are found. Because so many of my dvd's are badly marked as originating from progressive/interlaced sources I can say this yadif crash is something that will effect other users.

I've never been too lucky w/ YADIF...in non-double framerate mode anyway(my AVS script can't affort 50fps

)...it looks very ugly, like some drastic posterizing on PAL 576i?

Gabest's MPEG2 decoder in auto mode looks way better, and yet it doesn't blend fields...I think

haruhiko_yamagata · 25th February 2009, 12:29

Quote:

Originally Posted by mark0077

Oh definitely usable but consistently wrong for every DVD I have... Its not just a few pixels off really you can see in the screen shots

If you are talking about

It's just rendering quality.
The code is written in mmx, I know it's wrong, but it's a bit hard to fix.

B.F. · 25th February 2009, 12:38

Deblocking filter doesn't work in AviSynth and VirtualDub filters.

And can you do a checkbox in the Postprocessing page like "always postprocess" ?
It will be nice to use postprocessing filters even with a codecs like h264 sometimes.

mark0077 · 25th February 2009, 13:21

Quote:

Originally Posted by haruhiko_yamagata

If you are talking about

It's just rendering quality.
The code is written in mmx, I know it's wrong, but it's a bit hard to fix.

Thanks for letting me know the issue anyways. Any further ideas on my other problem, ie crashes in one frame areas when using external decoder or is that too tricky to work around?

haruhiko_yamagata · 25th February 2009, 15:06

Quote:

Originally Posted by mark0077

Thanks for letting me know the issue anyways. Any further ideas on my other problem, ie crashes in one frame areas when using external decoder or is that too tricky to work around?

What do you mean by "crash"? Is it memory access violation?

mark0077 · 25th February 2009, 15:16

Sorry i should say freeze. In the images above, if I use mpc-hc as the mpeg2 decoder, the menu looks perfect, but if I highlight one option, all is ok, but if I move the mouse up to another option, nothing happens. Media player classic still responds but no new frames are drawn so it appears as if the software has frozen. Its like ffdshow won't draw any subsequent frames when I highlight different options. This makes many dvds unusable without navigating through the player titles until the right one is found to start the movie.

MPC used on its own without ffdshow at all is ok, I can move the mouse around between options in the menu and no freezes. With ffdshow in the chain (but not decoding), I can only highlight one menu option, after that, no more screen updates happen and I will need to restart mpc to get back to a usable state.

iron2000 · 26th February 2009, 07:44

There are the edge-enhancement and denoise sliders in CCC, do they just work or need there be some ATI codec?
Or does it work only in DXVA? I'm confused of the video card's roles in playing my videos.
Don't seem to see a difference with those compared to the sharpen filter in ffdshow.

roozhou · 26th February 2009, 17:23

Hi, I have a question on ffdshow's deinterlacing filters.
When I use deinterlacer from DScaler in ffdshow, it always doubles the framerate. IIRC some of DScaler deinterlacers run in adaptive mode, that is doubling framerate at high-motion scene and weaving at low-motion scene. I also checked the DS_Deinterlace.h and there is

Code:

// Deinterlace functions return true if the overlay is ready to be displayed.
typedef BOOL (__cdecl DEINTERLACE_FUNC)(TDeinterlaceInfo *Info);

But ffdshow just ignores the return value and shows the frame always. Does ffdshow have some limitation on adaptive deinterlacing?

Konrad Klar · 26th February 2009, 20:19

If I change state of video codecs WMV1/7, WMV 2/8, WMV 3/9 to "disabled" MPC and MPC-HC crash when opening .wmv file.
This happens since rev2715, both in 32bit and 64bit version and for all .wmv files I have tested. WindowsMediaPlayer is not affected.
rev2698 and earlier are free from this problem (both 32bit and 64bit).
Example crash info:

Code:

Nazwa zdarzenia problemu:	APPCRASH
  Nazwa aplikacji:	mplayerc32.exe
  Wersja aplikacji:	1.2.1004.0
  Sygnatura czasowa aplikacji:	49a1f286
  Nazwa modułu z błędem:	libmplayer.dll
  Wersja modułu z błędem:	0.0.0.0
  Sygnatura czasowa modułu z błędem:	499c05b2
  Kod wyjątku:	c0000005
  Przesunięcie wyjątku:	00066cd5
  Wersja systemu operacyjnego:	6.0.6001.2.1.0.256.1
  Identyfikator ustawień regionalnych:	1045
  Dodatkowe informacje 1:	fd00
  Dodatkowe informacje 2:	ea6f5fe8924aaa756324d57f87834160
  Dodatkowe informacje 3:	fd00
  Dodatkowe informacje 4:	ea6f5fe8924aaa756324d57f87834160

I'm on:
Vista Ultimate x64 SP1 (polish).

25th February 2009, 15:16	#6757 \| Link
mark0077 Registered User Join Date: Apr 2008 Posts: 1,106	Sorry i should say freeze. In the images above, if I use mpc-hc as the mpeg2 decoder, the menu looks perfect, but if I highlight one option, all is ok, but if I move the mouse up to another option, nothing happens. Media player classic still responds but no new frames are drawn so it appears as if the software has frozen. Its like ffdshow won't draw any subsequent frames when I highlight different options. This makes many dvds unusable without navigating through the player titles until the right one is found to start the movie. MPC used on its own without ffdshow at all is ok, I can move the mouse around between options in the menu and no freezes. With ffdshow in the chain (but not decoding), I can only highlight one menu option, after that, no more screen updates happen and I will need to restart mpc to get back to a usable state. Last edited by mark0077; 25th February 2009 at 15:25.

26th February 2009, 07:44	#6758 \| Link
iron2000 Questioning User Join Date: Jan 2005 Posts: 166	There are the edge-enhancement and denoise sliders in CCC, do they just work or need there be some ATI codec? Or does it work only in DXVA? I'm confused of the video card's roles in playing my videos. Don't seem to see a difference with those compared to the sharpen filter in ffdshow. Last edited by iron2000; 26th February 2009 at 07:48.

26th February 2009, 17:23	#6759 \| Link
roozhou Registered User Join Date: Apr 2008 Posts: 1,181	Hi, I have a question on ffdshow's deinterlacing filters. When I use deinterlacer from DScaler in ffdshow, it always doubles the framerate. IIRC some of DScaler deinterlacers run in adaptive mode, that is doubling framerate at high-motion scene and weaving at low-motion scene. I also checked the DS_Deinterlace.h and there is Code: // Deinterlace functions return true if the overlay is ready to be displayed. typedef BOOL (__cdecl DEINTERLACE_FUNC)(TDeinterlaceInfo *Info); But ffdshow just ignores the return value and shows the frame always. Does ffdshow have some limitation on adaptive deinterlacing?

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

23rd February 2009, 20:01	#6745 \| Link
tetsuo55 MPC-HC Project Manager Join Date: Mar 2007 Posts: 2,317	Shouldn't the relevant information be passed on between each step in the filter graph? Information like: -AR(in all its flavours) -Framerate -Interlace flag Maybe even more information is usefull, if any value is no longer true(in the case of deinterlacing) the part of the chain that caused the change in the value should update the information to the new status and pass that along.

24th February 2009, 16:24	#6751 \| Link
flanger216 Registered User Join Date: Mar 2004 Posts: 140	This has been bugging me for a while: If you use libavcodec to decode MP3s, and you have 16, 24, 32 and 32-bit floating-point enabled as outputs, then ffdshow outputs 16-bit. If you deselect 16, 24 and 32-bit integer as output formats, then libavcodec outputs to 32-bit floating-point. So in the latter case, is libavcodec truly decoding to 32-bit floating-point, or is it still decoding to 16-bit and then merely padding the output to 32-bit floating point? This also applies to AC3, DTS, Vorbis and all the other lossy (bitdepth-less) codecs that libavcodec can handle. I only ask because libac3, libdts and libmad all seem to output to the highest bit-depth available, whereas libavcodec seems to prefer using the lowest available.

25th February 2009, 12:38	#6754 \| Link
B.F. Registered User Join Date: Jul 2004 Location: Russia, Novosibirsk Posts: 92	Deblocking filter doesn't work in AviSynth and VirtualDub filters. And can you do a checkbox in the Postprocessing page like "always postprocess" ? It will be nice to use postprocessing filters even with a codecs like h264 sometimes.