ffdshow tryouts project: Discussion & Development

[Leak]

Originally Posted by Reimar

It is not in any way important to me, but I am curious why it is impossible? libavcodec itself does offer the functionality, at least for some codecs.

Well, since all non-I-frames in MPEG2 fully depend on neighbouring decoded frames you can't just "decode it to half the size" and expect something watchable to come out of it.

And since you have to decode it at full resolution scaling it down afterwards only increases CPU load - and you can already do that using ffdshow's Resize & Aspect filter, anyway.

[Kurosu]

Originally Posted by Leak

Well, since all non-I-frames in MPEG2 fully depend on neighbouring decoded frames you can't just "decode it to half the size" and expect something watchable to come out of it.

And since you have to decode it at full resolution scaling it down afterwards only increases CPU load - and you can already do that using ffdshow's Resize & Aspect filter, anyway.

This is somewhat funny: would you also explain to David Huffman, were he alive, what variable length coding is?

Distortion is indeed a side effect, but it does improve decoding speed. The mplayer man page for instance, and google, have explanations about the lowres decoding option found in lavc.

[Leak]

Originally Posted by Kurosu

This is somewhat funny: would you also explain to David Huffman, were he alive, what variable length coding is?

Errr... so you're implying that David Huffman is to Variable Length Coding as Reimar is to MPEG2? (Sorry, I really genuinely don't understand that comment of yours...)

[LoRd_MuldeR]

Originally Posted by Leak

Errr... so you're implying that David Huffman is to Variable Length Coding as Reimar is to MPEG2? (Sorry, I really genuinely don't understand that comment of yours...)

I think what he actually meant is:
You don't need to explain how MPEG-2 decoding works to Reimar, because as an MPlayer/Codec developer he should be very familiar with it...

[Leak]

Originally Posted by LoRd_MuldeR

I think what he actually meant is:
You don't need to explain how MPEG-2 decoding works to Reimar, because as an MPlayer/Codec developer he should be very familiar with it...

Ehehe... ummm, since he asked that question above I didn't consider that he could be an mplayer dev...

Thanks for clearing that up...

[Mr VacBob]

Originally Posted by clsid

For libavcodec's mpeg-2 decoder there already is a "use speedup tricks" option in ffdshow. I don't know what that does precisely.

It (CODEC_FLAG2_FAST) uses a simpler dequantization step which results in more zero DCT coefficients. It's not that much faster, really.

What he asked for is controlled by AVCodecContext.lowres, which certainly works for MPEG2. I think it might already be exposed in the UI too.

Hi all,

There's a bug in ffdshow (reported at http://sourceforge.net/tracker/index...41&atid=867360 ) that makes some of the software I'm running unusable. I was just wondering if this bug was fixed in any of the nightly builds? If not, is it something that might be fixed in the future?

Thanks.

[leeperry]

is it me, or is it only possible to resize horizontally in ffdshow ?!

[jmartinr]

Originally Posted by leeperry

is it me, or is it only possible to resize horizontally in ffdshow ?!

It's you . Horizontal and vertical resize works fine here. Maybe the problem is in your renderer.

[leeperry]

well you can keep the exact same width and only resize vertically ?

now how do you do that ?

I've got a profile to resize to 1024x768, yet sometimes it resizes to 1152x768 with SD content.....I want freaking 1024*xxx

[cyberbeing]

It works here as well. Use "Specify horizontal and vertical size" > only change the vertical size > make sure Aspect ratio is set to "no aspect ratio correction".

I'm assuming you're using YV12 and YUY2 with VMR/EVR? With those color spaces VMR/EVR will sometimes use odd internal resolutions (which seems to be a built-in restriction) yet it will still display at your desired resolution.

[clsid]

Maybe those are anamorphic encodes? Check your PAR settings. Because it should work.

[JarrettH]

think we'll have a new beta soon?

[clsid]

At the end of the month if no regressions are reported.

[LoRd_MuldeR]

Originally Posted by clsid

At the end of the month if no regressions are reported.

Apparently the "Cubic Blending" deinterlacer doesn't deinterlace at all or it fails horrible to remove combing artifacts (which I think isn't possible with that method)

[Yong]

Guys, is it possible to make "custom" OSD display as list?
if the video resolution is small, OSD will got truncated if display all in one line like this


i hope this feature request isn't too hard for devs

[mike7]

Originally Posted by Yong

Guys, is it possible to make "custom" OSD display as list?

use \n for new line, ex.: %idimar\nRemaining time: %rtime

[Reimar]

Originally Posted by Leak

Ehehe... ummm, since he asked that question above I didn't consider that he could be an mplayer dev...

Thanks for clearing that up...

Well, you started from the assumption that decoding at half resolution is not possible in a useful way. But the reality is, that even quarter resolution decoding is possible with (IMO) acceptable quality losses simply by setting a libavcodec flag. Half resolution in my tests reduces decoding CPU usage by 60%, quarter resolution by another 20% (compared to half res) for high bitrate HD material.
So my question was about the part I did not know: why it would not be possible to export this functionality via ffdshow (seems I was mistaken to assume you knew about this libavcodec feature).

[leeperry]

Originally Posted by LoRd_MuldeR

Apparently the "Cubic Blending" deinterlacer doesn't deinterlace at all or it fails horrible to remove combing artifacts (which I think isn't possible with that method)

yeah the deinterlacing section of ffdshow is very poor.

the "cubic blending" doesn't do anything whatsoever, and the "linear blending" is very blurry.

the best deinterlace I've found so far is the PS "blending" script in MPC...but it only works w/ EVR or VMR9

Seb.26 told me that it's using a very simple script, but because noone is willing to add new features to ffdshow......it's not gonna happen anytime soon.

ffdshow is so great, but it's got a few annoying bugs that noone will ever fix I guess

for instance....when using Avisynth scripts, very often Reclock says "no video stream found", I stop/press play again, and it's fine...

[clsid]

but because noone is willing to add new features to ffdshow......it's not gonna happen anytime soon.

Comments like this ain't gonna help in upping the motivation of people. Mumbles something about free, open-source, spare time. Currently we only have the manpower to maintain ffdshow. For actual development we simply need a truckload of programmers with no personal life.


I assume that those deinterlacing problems exist for a long time already, and they are no recent regressions. So they won't hold back a beta. But anyway, I am no C++ programmer, and I don't really care about deinterlacing functionality.