madVR - high quality video renderer (GPU assisted)

[dansrfe]

Originally Posted by shaolin95

Hello guys...this is the first time I have tried madvr and it blew me away!
I have been using MPC for a long time with ffdshow, avisynth scripts etc... I have to try again today but unless my eyes were deceiving me yesterday, the quality is amazing! The color, detail , everything looks super.
I am planning to read a lot of this thread but one thing but I was wondering if there is a recommended card or cards list?
I only do 1080p bluray ripped to my main server and unless I am blind , this gives better quality than my Samsung player.
So, I have a HD2400Pro (maybe HD2600Pro need to verify) and this is only for HTPC so I am trying not to break the bank of course.
Any recommendations?


PS The reason I ask is because, I am not sure if its a card issue but it looks more like a sync issue where it is playing very smoothly then suddenly it hiccups and then goes fine again. My projector is running at 48fps (47.9xx) and I tried with Reclock as well. It apparently could be just at the beginning or for some movies only but I did this too late so couldnt try a lot. I will have to verify. I had to use DXVA (native) for this card or it would be a stutter fest.

The recommended card, as always, is usually the most expensive card that you are willing to pay for. The more you give to madVR, the more it can utilize.

[FlashGordon]

Some of my PAL DVDs are being detected as being SMPTE-C by madvr even though madvr also reports the resolution to be 720x576. I thought the primaries were determined by the resolution?

The strange part about this is if i play one of the individual VOBs with LAV Video as my decoder, madvr reports that it is using EBU primaries. If I use ffdshow libmpeg2 decoder and YV12 output and play the full DVD (with menus etc) madvr guesses it is SMPTE-C. Is there a reason for this?

EDIT: Forgot to test the individual VOB with ffdshow. madvr uses EBU when I do this as well. So this seems to only happen when I open the full DVD and it starts at the menu screen. Shouldn't madvr use EBU if it is still reporting a 720x576 resolution?

[Andy o]

Originally Posted by Pat357

No, second time still crashes and it seems it's not because of refresh-rate switching.
The crashing happens also if both files have the same fps, like 2 files with 23.970 fps.
Thanks for the tip about D3D11, but doesn't seem to have any influence on the crashing for me.
Did it completely solve this crashing problem for you ?

The other thing mentioned in this thread "flush" for "after copy to backbuff" + flush for "after last rendering step", seems to help, but does not solve the problem.
It just happens less likely

Yes, D3D11 works all the time for me, as does playing a second time after refresh change (but that of course is useless, though still fixes it). IIRC there was someone else having a similar problem, I just forgot about it for a while. My settings are pretty much default except for D3D11 and the refresh rate changer.

[TheShadowRunner]

Hi madshi,
Here are 2 small bugs in v0.81 that didn't happen with previous versions.
If you switch resolution manually, but stay under the same frequency, madVR doesn't go in exclusive mode anymore unless you reload the media.
Exemple: screen is on 1280x720@60. I start playback of a 1080p 30fps video and change the resolution manually to 1920x1080@60. In this case playback of the media continues just fine but madVR won't go in Exclusive mode anymore. Just reloading the media at this point allows madVR to go in Exclusive mode again.
I know you don't really like people switching resolution "behind madVR's back" but it now works perfectly fine for all the other cases, and this used to work fine with 0.77-0.80 too.

The 2nd bug, this time directly related to madVR's own display changer.
I put only 1080p60 in the "list all display modes madvr may switch to". I put the screen in 720p50 manually. I then start playback of a 1080p 30fps video, in this case madVR doesn't switch the resolution to 1080p60. This also used to work fine with 0.77-0.80.

Absolutely everything else runs flawlessly here, a real pleasure

Edit: "delay switch to exclusive mode by 3 seconds" is enabled.
Btw, it would be great if we were able to tweak this value, in milliseconds, for testing, maybe via registry key?

[shaolin95]

Originally Posted by dansrfe

The recommended card, as always, is usually the most expensive card that you are willing to pay for. The more you give to madVR, the more it can utilize.

So I noticed something...my 2600Pro is able to run the 1080p movies great if I uncheck the option to go to exclusive mode when going to full screen then there are no hiccups, the movie runs smoothly all the way. I wonder what that option does and what am I missing..
Any ideas?

[Plutotype]

Originally Posted by Andy o

Yes, D3D11 works all the time for me, as does playing a second time after refresh change (but that of course is useless, though still fixes it). IIRC there was someone else having a similar problem, I just forgot about it for a while. My settings are pretty much default except for D3D11 and the refresh rate changer.

When playing a video in exclusive mode with D3D11 and right-clicking to access madVR edit options, it pauses the video sometimes. Also I get presentation glitches with D3D11, one always just after playback starts and couple of them appear also later.

That is why I use D3D9 with the option "overshoot max frame latency". Results are 0-1 dropped frames, 0 presentation glitches. Of course on my setup.

[madshi]

Ok, easy answers first. Replies which require some checks will come a bit later.

Originally Posted by ryrynz

MPC generates a error saying it's stopped working, so I'm uncertain, I've created a log anyway if you're interested.

http://www.mediafire.com/?5a6o4g47fulerfs

A log alone won't help much for a crash. Was the crash in MPC-HC or in madVR? If it was in madVR you should (ideally) have seen the madVR exception box with an option to send a bug report to me. If you just saw the standard windows crash box then the crash was probably in MPC-HC and not in madVR. Without a madVR crash report there's not much I can do about crashes.

Originally Posted by sneaker_ger

This is unrelated to the new release as I have the same problem with 0.80, but does anyone else also have problems getting deinterlacing to work?
I'm on Win 7 x64, Radeon HD 5850, Catalyst 12.2 and deinterlacing just won't work, even if I force it via ctrl+alt+shift+d, I still see all those combs. It worked fine when I used it a few weeks ago (or use a different filter like LAV Video to do the deinterlacing), so maybe it is something with the new drivers?

Originally Posted by glc650

I can't get deinterlacing to work either. When I turn on deinterlacing in madVR (and turn off software deinterlacing in LAV Video Decoder) and try to play back my 1080i60 content I get audio but no video (video is black/blank in mpc-hc). Had the same issue with version .80. I've got a Radeon 6570 w/Catalyst 12.1.

A few things to check:

(1) Press Ctrl+Alt+Shift+D to make sure deinterlacing is turned on.
(2) Press Ctrl+Alt+Shift+T to check whether madVR is in film or video mode.
(3) Double check whether deinterlacing works with other renderers (VMR9, EVR).

Originally Posted by ajp_anton

There are some weird subtitle behaviour when there are media player messages in the top left corner. I guess this is why you've disabled the "pause" messages, but if you seek, change volume etc, the problem is still there. And I'd like to have the pause message =).

That's technically not possible to achieve at the moment. When a message appears on screen, madVR is forced to rerender the frame. In order to do that, the subtitles must be rerendered, too. And the internal MPC-HC subtitle render engine doesn't like to go backwards in time. The going backwards happens because madVR has already asked the subtitle renderer to render future frames in order to fill the render queue. So in order to rerender the "current" frame, the subtitle renderer would have to "rewind". And whenever I ask it to do that, it needs a couple of frames until it outputs subtitles again. As a result, subtitles disappear. There's not much I can do about it atm. There may be a way in the future to solve this, by using a better subtitle renderer, but that doesn't exist yet.

Originally Posted by TheShadowRunner

And from what I've tested so far, everything is perfect; thank you very much!

Originally Posted by THEAST

Hmm, using v0.80, my 3-year old HD4870 literally died trying to play the Life in the Garden 4k resolution video; it was totally unplayable! But with the new version, the video plays semi-fine with 1-2 dropped frames per second. Great performance increase here.

Good to hear!

Originally Posted by JarrettH

bcvbc

Please quote the "normal" way. I've had to manually copy your texts over because the doom9 forum software doesn't understand your way of quoting/replying.

Originally Posted by JarrettH

Yeah, I mean it looks better with madvr's IVTC instead of dxva deinterlacing. I'm not sure if there's more going on technically, but in the case of this DVD it wanted to use deinterlacing first.

That's kinda as expected. Currently madVR is not able to autodetect whether a DVD is "film" or "video" type. So it defaults to DXVA deinterlacing because that can handle everything. If you know for a fact that IVTC will work on any given DVD, just turn it on manually and you may get better quality (at least better smoothness). But if you turn IVTC on for material that can't be IVTCed, you'll get combing.

Originally Posted by JarrettH

Yes and Windows 7 with a GTX 550 Ti 285.62 WHQL drivers

I'm aware of that there are sometimes odd effects with the D3D11 presentation option. I plan to work on that in a future version. But it won't be easy to fix, because it only rarely happens on my PC.

Originally Posted by Pillars of Creation

Using madVR 0.81

Frame rate                       : 25.000 fps       <----
Original frame rate              : 23.976 fps       <----

will use original frame rate (23.976 fps "says source filter" with dropped frames) instead of frame rate (25.000 fps).

Adding .25p at the end before .mkv will play file with 25.000 fps (without dropped frames)

It is not madVR's job to look at the container information. That's the splitter's job. The "says source filter" is the information that the splitter or decoder provided madVR with. So if that information is incorrect, you need to complain to the splitter/decoder developer.

Originally Posted by Seegs108

Has anyone noticed that MadVR doesn't work properly on a monitor in portrait mode?

Does every GPU/monitor support that? I don't even know how to activate that?

Originally Posted by chros

About presentational glitches, I found out what caused the problem earlier for me on the laptop: tha LAN cable !

Ouch.

Originally Posted by chros

Well, not totally solved:
- after reboot playback is awful again (many-many glitches, and frame drops)
- but unistalling and reinstalling the same new lan driver solves the problem again (very few glitches or none) !

So I don't know what is going on ...

I don't know, either. You could try checking if you can change something in your BIOS. In any case, if reinstalling the LAN driver makes the problem go away then most probably madVR is not at fault.

Originally Posted by dansrfe

Wow, I can now FINALLY drag MPC-HC between screens and fullscreen to exclusive mode without lag and without a frozen screen. Before exclusive mode could only be started on one screen and then I would have to reopen the video clip to use fullscreen exclusive mode on my other screen.

I must say that Windowed <-> exclusive mode and rapid switching between exclusive modes of different screens is now completely smooth and fixed from my viewpoint.

Good to hear. I hope it will stay this way once the refresh rate changer bugs are fixed.

Originally Posted by dansrfe

Feature request: Is it possible to add key commands to enable (or invoke) the display changer and to disable (or stay at the original refresh rate)? Basically to be able to have the display changer 'execute' at will?

Possible, yes, but I'm not sure how many users would actually use that. I'm a bit reluctant to add keyboard shortcuts for any funny feature. The alphabet has only so many characters.

Originally Posted by mark0077

Hi madshi, today in mpc-hc I got a message about mpc-hc crashing but I see a madVR crash report was created on the desktop. Is it useful to send this to you somehow?

Yes, that's why it was created.

Originally Posted by magilvia

I vote for a fix to the "DVD macromedia fail" on window seven, if you please.

It's on my to do list, together with so many other things.

Originally Posted by mbordas

is dscaler5 still necessary with madVR 0.81? I'm a little confused whether the ivtc features in madVR duplicate what dscaler does for dvds (mpeg2).

In theory madVR's IVTC should be better than the DScaler5 IVTC Mod in every way. Whether that's true in real life is another question and still needs to be proven over time.

Originally Posted by mrcorbo

I'm finding that the automatic display mode changer isn't taking deinterlacing+IVTC into account when determining which display mode to change to. It is selecting 1080p60, even though the playback rate is 24fps. When the content is already 24p, 1080p24 is selected correctly. Can anyone confirm, or am I missing something?

Did you force madVR into film mode? Only in film mode madVR is supposed to switch to 24p, when playing back a 1080i60 file.

Originally Posted by Zachs

Would it be possible to change the seek bar's height in exclusive mode? It's a little too thin now -- with an overscan, I could merely see the tip of the bar.

A time code overlayed on top of the seek bar (I know it's custom drawn and much harder than what most ppl think) would be appreciated as well.

Yeah, that would be nice, but just changing the seek bar height alone isn't enough, I'd also have to redraw the seekbar thumb image. And adding a time code overlay would also cost extra work. This will surely all come sooner or later, it's just a question of priorities. Everything which can't be done in a couple of minutes has to wait until it's the most important thing in my to do list.

Originally Posted by turbojet

Thanks madshi for the new version, bicubic 100 exceeded my expectations and unlike EVR's bicubic 100 there is very little aliasing and less ringing.

Glad you like it. Not sure why it has less ringing than EVR's algorithm, though. Ok, my algorithm should be mathematically "perfect". Maybe EVR's algorithm isn't? I don't really know...

Originally Posted by MokrySedeS

Speaking of FSE seek bar, it would be nice if it supported seeking to keyframes with Shift+LeftClick (or some other way) just like MPC-HC does.

Does anybody know how to do this programatically? I don't right now from the top of my head. "IMediaSeeking" allows me to seek to a specific time with the hint to the decoder to seek to the nearest key frame. But I've no idea how to find out which time code the next key frame might have.

Originally Posted by iron2000

On further testing it seems to be the enabling of 'delay playback start until render queue is full' and probably a high backbuffer value.

That combination may explain a certain delay in switching. But it doesn't explain why you didn't have the same problem in v0.80, so I'm a bit confused.

Originally Posted by Mixer73

Got a wierd symptom with MPC-HC, LAV Filters & madVR 0.81, black screen with the above using RGB32, but works right away in YV12.

Can you reproduce that? Does it happen with all videos or just with some specific ones? In the latter case can you upload a small sample? If it happens if all videos, something is weird. Maybe a small madVR debug log would help, not sure.

Originally Posted by italospain

I have a request:

madVR CPU resizers.

Maybe at some time in the future, but not anytime soon. Creating fast CPU resizers would probably require carefully written SSE2 code which is a quite time consuming thing to do.

Originally Posted by pirlouy

About my previous issue:
I've tested with 0.81: no problem
I came back to 0.80 : no problem

So, it was on my own. Sorry for the disturbance. :/

No problem. There was a reason I never analyzed that problem. It didn't seem likely to me that madVR would be at fault.

[DragonQ]

When using MadVR with LAV set to deinterlace 50i to 50p, it says "movie 25.000 fps (says source filter)", even though it's definitely playing back at 50p. Is this a bug?

[madshi]

Easy answers, part II:

Originally Posted by mbordas

Follow up to my question about dscaler5:

If I play this clip:

https://rapidshare.com/files/4264372196/elizabeth.vob

with dscaler5, madVR correctly reports the movie is 23.976 fps and sets the TV to that refresh rate.

without dscaler5, I can press CTRL-ALT-SHIFT-T to force it to set the display to 23.976, but it won't do so automatically unless I disable automatic detection and select "default to film mode".

Seems like dscaler is still a better option for the time being?

What you're describing is the current intended behaviour. madVR is currently not able to autodetect whether a video file is "film" or "video" type, so it always defaults to "video" type because that results in DXVA deinterlacing and DXVA should be able to handle all kinds of materials. If you only play movie files and no native video content (music concerts, sports) at all, then you can simply force madVR into "film" mode. Then it should work identically to (or better than) DScaler.

Originally Posted by mbordas

And in that case it still reports the movie is 29.970 fps.

No. It just informs you about what the splitter/decoder is telling madVR, nothing else. The splitter/decoder is sending madVR 1080i60 content, so the 29.970 information is absolutely correct. It's only after madVR's IVTC that the movie is converted from 1080i60 to 1080p24. So practically the input is 1080i60 (and 29.970 is the correct information for that) and the output is 1080p24.

Originally Posted by mbordas

why then is it reported as 29.970, when mediainfo says it's 23.976?

Because mediainfo tries to be clever (and sometimes fails with that). mediainfo tries to tell you which frame rate the video file has after IVTC. But madVR gets the video frames before they're IVTCed, so it's correct for the splitter/decoder to report 29.970.

Originally Posted by robpdotcom

How does madVR's ivtc filter handle a file like this: http://www.mediafire.com/?2obecsdjb930dgg

The file is 1080i59, and according to tsmuxer, has pulldown flags. While playing the file, Reclock shows 23Hz (and when using EVR, mpc also shows 23fps). I use LAVvideo, which I assume must be stripping the pulldown.

However, madVR shows 29Hz, and shows a 3:2 cadence.

The file plays fine, so I assume everything is working right, but I didn't think the two things above could be true at the same time.

Reclock, EVR and MPC don't take "repeat first field" into account. They just count the number of frames that are sent to the renderer. If you have a file which switches multiple times between soft-telecine and hard-telecine (many broadcasts do), Reclock, EVR and MPC will all switch betwen 23Hz/fps and 29Hz/fps, as well, or even show some random value betwen 23 and 29. In contrast madVR properly interprets "repeat first field" and thus it doesn't switch between 23 and 29. Instead madVR always correctly reports 29 with a 3:2 cadence. Which is the correct information.

Originally Posted by Andy o

I'll be damned, I completely misread that option. Checking disables it, now that makes more sense considering my previous post to madshi about it! You're completely right. Sorry for the confusion.

Now I'm not sure: Is your previous post still valid?

Originally Posted by janderclander14

I've recently found an annoying problem. Whenever I play any 50fps (progressive) video, MadVR crashes within 1 or 2 minutes of start playing (mainly in exclusive mode). This does not happens when playing 23.976 or 60 fps videos. In the former case, statistics show a composition of 59.9 fps instead of 50.

I've tried other renderers (EVR) and videos does not crash. The problem persists whenever I disable/enable Reclock or I change options in MadVR like queue lengths and happens with any 50fps content.

I use the latests versions of MadVR, LAV filters, ffdshow, MediaPlayer HC, and Reclock.

Is it really madVR that crashes? If so, with v0.81 you should get a madVR crash box with the option to send a bug report to me. Please do so. If you get a standard windows crash box, then maybe the crash didn't occur in madVR but somewhere else? In any case, if you're overclocking your PC, try running at stock clocks.

Originally Posted by VHT

Question about dithering.Seems that latest LAV Video decoder has dithering option and it can't be ticked off, so should I set dithering off from madVR instead or let them both do dithering?

Never turn off dithering in madVR, unless you have good reasons for that. The only good reasons that come to my mind are:

(1) If it's the one and only chance to get smooth playback, after you've already dumbed down all other settings.
(2) If you want to check whether screenshots are bitwise identical.

Originally Posted by kerman

madshi, given most of the media are 16-235, wouldnt it be more accurate to the source for displaying, to setup tv to limited RGB (16-235) and madvr to TV levels?

If so, how would you configure coreAVC as for Output and Input levels? (0-255), (16-235) or Auto detect. I think you said you use coreavc for h264 decoding so just curious how you have set it up. Im going crasy to be sure to have all the luminance configuration filters correctly set up

Originally Posted by nevcairiel

i would recommend to set input and output in CoreAVC to 16-235, that'll ensure it doesn't touch the values (independent of what you want to output later)

^

Setting madVR to TV or PC levels output is kinda difficult to recommend. Setting it to PC levels means that you can calibrate your display accordingly and still get a proper grayscale for games and desktop. So that might be the best solution. But your display must be able to handle that.

Originally Posted by 703

I have a .mp4 file that is tagged "Matrix coefficients : BT.709-5, BT.1361, IEC 61966-2-4 709, SMPTE RP177 " However, MadVR is still "best guessing" as BT.601? I'm using Lav Decoders within MPC-HC.

Tagged how? In the file name? madVR currently does not read matrix coefficients from the file name. Or do you mean it's tagged in the video headers? Or in the container? If it's in the video headers then madVR should be able to identify it (only if it's h264, MPEG2 or VC-1). Otherwise, if it's in the container then it's the splitter's job to forward this information to madVR. In that case talk to the splitter developer.

Originally Posted by kasper93

I bought DVB-T tuner on chip IT6135 and unfortunately it isn't working with madVR as renderer in MPC-HC. With EVR or Haali Renderer works good, but with madVR it fails to load. I know that there are more important things to change in madVR, but if you could add this to your to do list, and someday maybe you will fix it

Log:
http://dl.dropbox.com/u/16282309/mad/madVR.DVB-T.7z

I don't think I can test this without having the DVB-T tuner hardware?

Originally Posted by shaolin95

Hello guys...this is the first time I have tried madvr and it blew me away!
I have been using MPC for a long time with ffdshow, avisynth scripts etc... I have to try again today but unless my eyes were deceiving me yesterday, the quality is amazing! The color, detail , everything looks super.
I am planning to read a lot of this thread but one thing but I was wondering if there is a recommended card or cards list?
I only do 1080p bluray ripped to my main server and unless I am blind , this gives better quality than my Samsung player.
So, I have a HD2400Pro (maybe HD2600Pro need to verify) and this is only for HTPC so I am trying not to break the bank of course.
Any recommendations?

PS The reason I ask is because, I am not sure if its a card issue but it looks more like a sync issue where it is playing very smoothly then suddenly it hiccups and then goes fine again. My projector is running at 48fps (47.9xx) and I tried with Reclock as well. It apparently could be just at the beginning or for some movies only but I did this too late so couldnt try a lot. I will have to verify. I had to use DXVA (native) for this card or it would be a stutter fest.

DXVA decoding is kinda troublesome for ATI cards when using madVR. So that may explain the problems you're having. If you can't upgrade your CPU in order to do software decoding then I would suggest getting a budget NVidia GPU (something faster than a 520, though), or maybe the new Radeon 7750. I'm not totally sure the 7750 will work perfectly with DXVA, but at least 1080p24 movies should be no problem.

But the first step should be to make sure that it's really a problem with the DXVA decoding. So I'd suggest that you activate the madVR debug OSD (Ctrl+J) and then have a look at all the various queues. In the moment when the image stutters the debug OSD should show an increase in "dropped frames". In that moment please check which of the queues are empty and which are not. The most important is the top most queue which is empty. Is it the decoder queue? Or a different one?

Originally Posted by FlashGordon

Some of my PAL DVDs are being detected as being SMPTE-C by madvr even though madvr also reports the resolution to be 720x576. I thought the primaries were determined by the resolution?

The strange part about this is if i play one of the individual VOBs with LAV Video as my decoder, madvr reports that it is using EBU primaries. If I use ffdshow libmpeg2 decoder and YV12 output and play the full DVD (with menus etc) madvr guesses it is SMPTE-C. Is there a reason for this?

EDIT: Forgot to test the individual VOB with ffdshow. madvr uses EBU when I do this as well. So this seems to only happen when I open the full DVD and it starts at the menu screen. Shouldn't madvr use EBU if it is still reporting a 720x576 resolution?

If the MPEG2 track has primaries information in the headers, madVR should be able to detect that. If it doesn't, madVR guesses based on the resolution. Maybe the menu screen is 720x480 and the movie later 720x576? Or maybe the menu screen comes with primaries = SMPTE-C information? I'm not sure from the top of my head whether madVR switches to different primaries if the video stream changes in the middle of the movie. Anyway, what does the madVR OSD say about the primaries? Does it say "best guess" or does it say "says bitstream" or something else? Also check the OSD to see the resolution of the menu screen.

Originally Posted by TheShadowRunner

Here are 2 small bugs in v0.81 that didn't happen with previous versions.
If you switch resolution manually, but stay under the same frequency, madVR doesn't go in exclusive mode anymore unless you reload the media.
Exemple: screen is on 1280x720@60. I start playback of a 1080p 30fps video and change the resolution manually to 1920x1080@60. In this case playback of the media continues just fine but madVR won't go in Exclusive mode anymore. Just reloading the media at this point allows madVR to go in Exclusive mode again.

Please try this again with the next build. It might be fixed there. If not, let me know.

Originally Posted by TheShadowRunner

The 2nd bug, this time directly related to madVR's own display changer.
I put only 1080p60 in the "list all display modes madvr may switch to". I put the screen in 720p50 manually. I then start playback of a 1080p 30fps video, in this case madVR doesn't switch the resolution to 1080p60. This also used to work fine with 0.77-0.80.

That's weird! Can I have a debug log from this situation, please?

Originally Posted by TheShadowRunner

Edit: "delay switch to exclusive mode by 3 seconds" is enabled.
Btw, it would be great if we were able to tweak this value, in milliseconds, for testing, maybe via registry key?

Not at this time. Too many other things to do.

Originally Posted by DragonQ

When using MadVR with LAV set to deinterlace 50i to 50p, it says "movie 25.000 fps (says source filter)", even though it's definitely playing back at 50p. Is this a bug?

The "says source filter" information is what the decoder is reporting to madVR. The decoder is outputting 50 fields per second, or 25 frames per second. So the information is correct. The 50p output is only due to deinterlacing which is done inside of madVR. So everything is correct.

[agustin9]

Hi madshi, can you make the statistics display the current frame rate like evr-cp?

[nevcairiel]

Originally Posted by madshi

Otherwise, if it's in the container then it's the splitter's job to forward this information to madVR. In that case talk to the splitter developer.

As if there is a way to communicate from the splitter up to the renderer. The decoder would swallow that information.
Really, most video codecs have fields for that in their header, doing that in the container is just backwards.

[DragonQ]

Originally Posted by madshi

The "says source filter" information is what the decoder is reporting to madVR. The decoder is outputting 50 fields per second, or 25 frames per second. So the information is correct. The 50p output is only due to deinterlacing which is done inside of madVR. So everything is correct.

I thought with the YADIF option enabled in LAV, the deinterlacing is being done by LAV itself, before MadVR gets the frames?

[madshi]

Originally Posted by agustin9

Hi madshi, can you make the statistics display the current frame rate like evr-cp?

Yes, I can. But I won't, at least not any time soon. It would cost me many hours to implement that. That time is better spent elsewhere, IMHO.

Originally Posted by nevcairiel

As if there is a way to communicate from the splitter up to the renderer. The decoder would swallow that information.
Really, most video codecs have fields for that in their header, doing that in the container is just backwards.

I agree that this information should be in the video bitstream. But we're not living in an ideal world, so we have to do the best we can do with imperfect files. From an end user point of view, if the container does contain primaries information and the video bitstream does not, then the information from the container should be used. From a DirectShow point of view this is perfectly possible. The splitter can output this information via VIDEOINFOHEADER2, the decoder can then forward this information to the renderer. Ok, no decoder today does that (except for the madVR internal decoders, haha, but that doesn't count, I guess). But hey, you're developing both a splitter and a decoder...

BTW, doesn't the same logic already apply to e.g. the aspect ratio and to the frame rate? E.g. if the container contains AR and FPS information and the video bitstream does not, doesn't the splitter output both and the decoder forward both? At least that's how it should be, IMHO.

Originally Posted by DragonQ

I thought with the YADIF option enabled in LAV, the deinterlacing is being done by LAV itself, before MadVR gets the frames?

With YADIF enabled in LAV you should see "50.000 fps (says source filter)". If you don't then you need to talk to nevcairiel.

[TheShadowRunner]

Originally Posted by madshi

Please try this again with the next build. It might be fixed there. If not, let me know.

[About manual resolution switching preventing Exclusive mode]
Will do, thank you!

That's weird! Can I have a debug log from this situation, please?

[The 2nd bug, this time directly related to madVR's own display changer.
I put only 1080p60 in the "list all display modes madvr may switch to". I put the screen in 720p50 manually. I then start playback of a 1080p 30fps video, in this case madVR doesn't switch the resolution to 1080p60. This also used to work fine with 0.77-0.80.]

Sure, here it is:
http://videoff7.free.fr/madVR_debug_TSR.zip
See you,

TSR

[agustin9]

Originally Posted by madshi

Yes, I can. But I won't, at least not any time soon. It would cost me many hours to implement that. That time is better spent elsewhere, IMHO.

Ok madshi, thanks for your hard work. With that and the repited frames counter madvr would be nearly perfect

[madshi]

Originally Posted by TheShadowRunner

I put only 1080p60 in the "list all display modes madvr may switch to". I put the screen in 720p50 manually. I then start playback of a 1080p 30fps video, in this case madVR doesn't switch the resolution to 1080p60. This also used to work fine with 0.77-0.80.

The log says that the upstream filter (splitter/decoder) did not inform madVR about the framerate of the movie file. As a result madVR doesn't know what to do. I don't see how 0.77-0.80 could have behaved any different in this situation. Please double check whether the problem really does not occur with 0.80. I rather think the video file or the splitter/decoder are at fault, and that the same problem would have occurred with 0.77-0.80, too.

Originally Posted by agustin9

Ok madshi, thanks for your hard work. With that and the repited frames counter madvr would be nearly perfect

A "repeated frames" counter is on my to do list, I consider that important (but not as important as some other things on my to do list).

[TheShadowRunner]

Originally Posted by madshi

The log says that the upstream filter (splitter/decoder) did not inform madVR about the framerate of the movie file. As a result madVR doesn't know what to do. I don't see how 0.77-0.80 could have behaved any different in this situation. Please double check whether the problem really does not occur with 0.80. I rather think the video file or the splitter/decoder are at fault, and that the same problem would have occurred with 0.77-0.80, too.

Thanks for checking it this fast.
I just tested with build 0.77 and you are right of course, it doesn't switch either.
I'm sorry

I double-checked the other bug at the same time (manually switching resolution prevents FSE if frequency stays the same, eg: 1280p60 -> 1080p60), it does work with 0.77 and broken with 0.81, no mistake on my end on this one ^^;

[agustin9]

Originally Posted by madshi

A "repeated frames" counter is on my to do list, I consider that important (but not as important as some other things on my to do list).

Sorry about that, not my native language

What are the next things on your list?

[shaolin95]

DXVA decoding is kinda troublesome for ATI cards when using madVR. So that may explain the problems you're having. If you can't upgrade your CPU in order to do software decoding then I would suggest getting a budget NVidia GPU (something faster than a 520, though), or maybe the new Radeon 7750. I'm not totally sure the 7750 will work perfectly with DXVA, but at least 1080p24 movies should be no problem.

But the first step should be to make sure that it's really a problem with the DXVA decoding. So I'd suggest that you activate the madVR debug OSD (Ctrl+J) and then have a look at all the various queues. In the moment when the image stutters the debug OSD should show an increase in "dropped frames". In that moment please check which of the queues are empty and which are not. The most important is the top most queue which is empty. Is it the decoder queue? Or a different one?

Thanks for the replay
I have a core duo 6600 at 4.2 but that is not enough I think plus I love how this works. Also I notice that the hiccups only happen when I am in Full Screen Exclusive mode. If I dont use Exclusive then full screen plays smoothly. Am I missing something by not using Exclusive.
If you feel an Nvidia card works better , I can do that, I dont mind either company.

Not sure if a newer card will have better IQ but just in case, for Nvidia do you have a preferred card for this only?

[madshi]

Originally Posted by FlashGordon

madshi, here are the samples you requested:

http://www.mediafire.com/?d9402uyc95gk48m

I just briefly went over the first sample with 0.81 and I actually see some combing in different spots. The second sample is from a PAL dvd and I use ReClock to play it at 24 fps, so I'm not sure if that affects the behavior of madvr's IVTC

Let me know if you need more/longer samples

Thanks - interesting samples! The first one ("whatever you do, dearie, don't get married") is the best. It's a pretty straight 4:3:3 cadence, but every once in a while there's a glitch in the encoded cadence. madVR detects that, but sometimes a bit too late, so that there is sometimes a bit of visible combing. This will be a good sample for later, when I try to improve the speed of detecting such encoded cadence errors. Can I please have a longer version of this video? The longer the better. Thanks!

The second sample (PAL) is weird. It's the only PAL sample with a 3:2 cadence that I've ever seen. madVR detects the cadence correctly, but the video is encoded rather poorly. There are some combing artifacts which are encoded into the source. Nothing madVR can do about.

The third sample is a typical case of field blended encoding. These can only be properly handled by video mode deinterlacing. IVTC won't work correctly on those.

Originally Posted by agustin9

What are the next things on your list?

I don't really like to talk about future development. I'll talk about it when it's done. Also, I don't have my to do list sorted. When I have time to implement something new, I look through the list and check what seems to be most important (to me).

Originally Posted by shaolin95

I have a core duo 6600 at 4.2 but that is not enough I think plus I love how this works. Also I notice that the hiccups only happen when I am in Full Screen Exclusive mode. If I dont use Exclusive then full screen plays smoothly. Am I missing something by not using Exclusive.
If you feel an Nvidia card works better , I can do that, I dont mind either company.

I think at 4.2GHz it should be fast enough for CPU decoding? It only hiccups in exclusive mode? That's weird. Anyway, if you get perfect results in windowed mode then there's no problem with disabling exclusive mode. I mentioned NVidia only because you wanted to use hardware decoding. Before changing GPUs I'd really suggest checking whether that's really the cause of the problem. If the problem only occurs in exclusive mode then it's probably not caused by hardware decoding. Summary: If you have no problems at all with fullscreen windowed mode, then there's no need to use exclusive mode, and no need to upgrade your GPU, either.