madVR - high quality video renderer (GPU assisted)

[Aktan]

Yes, but the DVD / .VOB doesn't have any such limitation. Two successive chapters could alternate from true interlaced content with a hard telecine encoded into it to p24 w/ soft pulldown and back. If the decoder ignores the soft pulldown and decides to output p24 content, what is it supposed to do when it gets to the next chapter that's true interlaced content?

I agree with Stereodude here huhn. I've seen content in Anime DVDs where most of the episode is FILM but the opening and ending English credits are overlay in VIDEO (go figure).

[huhn]

let's take your sample as an example.

this file is not flagged interlaced so how should it switch to interlaced with it? madVR IVTC wouldn't work at all. if it would send now 3:2 IVTC could still see it and fix it. so what has changed

if you play a DVD directly from the disc without a remux than this is a different story.

but if you want to be always save you have to use deitnerlacing always and never IVTC.

[chros]

searched again and found but that are 8 versions on the site, thanks a lot! but i still don´t know what version is my...
http://www.panelook.cn/modelsearch.php?panel_part_number=B156HW01
It doesn't really matter: they are all 6bit+FRC :)

i tried many times make a 3dlut but not achieve any descent result, all are too saturated, darker with banding, tried many configurations but nothing

I have a computer which I use for amature photo editing at home and it's calibrated with i1 Display pro and basICColor (it generates an icm file which is loaded by Windows at startup) Just today I found out about dispcalGUI and Argyll CMS. Since I also watch movies on this computer, what's the advantage of dispcalGUI over my existing calibration besides the 3D LUT? Is the current movie playback with MadVR using the .icm of the existing calibration?
3dlut gives you more accurate colors, there's a good article about this here: https://web.archive.org/web/20150213213837/http://www.lightillusion.com/luts.html
But let's continue the discussion here: http://forum.doom9.org/showthread.php?t=172783

[dbcooper]

Is there anyway to use the AMD Catalyst "video enhancements" with MadVR in MPC-HC? I like to use a little bit of their de-blocking (10) and mosquito noise reduction (5) with lower quality sources. It works with EVR/LAV Video (dxva native), but not with MadVR. Any suggestions?

[huhn]

should work with DXVA scaling.

[Stereodude]

Or if you're using deinterlacing.

[madshi]

That is another log. Thanks!
https://drive.google.com/file/d/0B7a6LffuxvKUSjJMWU1pQ0QtR3c/view?usp=sharing
Ok, I've once again tried to fix it. Let's see if it works now in the next build.

Do typical consumer HDTVs and projectors actually support DDC/CI, anyway? Even computer monitors rarely support it, except maybe professional ones.
I've no idea.

Do my calibration settings in dispcalgui i thought the correct was use the already calibrated to bt 709 with pure power curve transfer function

but testing the hdr content i only achieve goods resuls if i set the bt 709/601 curve

could someone explain the difference between the the two transfer functions?
Simply two different types of transfer functions (gamma curves). Normally, if your display is calibrated well, you should choose the transfer function your display is calibrated to.

This is raw YUY2 from an analog capture source (device). I don't think it can deliver frames faster than realtime as it captures in realtime.
Oh well, in that case I'm not sure if madVR can work well at all with this kind of source. I mean live TV sources somehow manage to fill the decoder queue. If the analog capture source can't, then there's not much I can do about it. Maybe I can try to optimize this in some far away (!) future build, but for now madVR expects the queues to be nicely filled. And for that the source needs to be able to deliver frames faster than real time. Maybe you could trick the source to do that by intentionally delaying audio by a full second or something like that?

Yes, I am using DXVA upscaling, but when I use any other upscaling with this video (using last versions of MPC-HC, XYSubFilter and MadVR) I don't get any image at all. The sound plays fine but the screen is black. With some of the upscaling methods I can see the info when typing Ctr+J, but in most cases I see nothing at all. All the upscaling methods work fine with other videos that I have tried.

How I can create the sample?
I took a sample with Avidemux but it ignored the subtitles. Then I extracte d the subtitles from the original video and I added them to the sample. But something is wrong. Selecting DXVA upscaling it was possible to reproduce the problem: subtitles not upscaling and appearing in a wrong position, but when I tried Catmul-Rom on the sample the computer crashed.
To me this sounds like there's some general problem with your PC, either with the software installation or the hardware. Or does this problem only occur with this one specific video file? Can you use Catmull-Rom upscaling for other videos without any problems?

I'm using MPC-HC 64bit nightly 1.7.10.28 from Dec 1st, and I've exported its settings to a file (http://wikisend.com/download/128016/mpc-hc-settings.7z) to make it easy to test.

in the Video Frame menu:
- Touch window from inside = selected
- Keep Aspect Ratio = enabled
- Override Aspect Ratio = Default

What I describe below is when using software decoding, and with madVR performing DXVA deinterlacing:
- the copyright warning shows, and in the OSD it says: deinterlacing off [says upstream]
- then the studio logo display starts, but for some reason the player moves from 00:00 to 00:01 to 00:02 to 00:03 seconds slower than real-time (it takes more than 3 seconds); while this happens, the OSD still says: deinterlacing off [says upstream]
- then the playback of the studio logo resumes to real-time, and the OSD says: deinterlacing on [says upstream]
- while the studio logo is displayed there is a moment close to its end when it switches to 16:9, stays like this for half a second, then switches back to 4:3 before the logo clip ends
- then the menu starts, and the display ratio is still 4:3; OSD still says deinterlacing on [says upstream]

The effect when using hardware decoding and deinterlacing performed by QuickSync is the same, just that the madVR OSD always says "deinterlacing off [says upstream]".

The settings used for madVR are the same as for the other movie clip. You can disable smooth motion, and it still happens. The TV refresh rate is 60Hz (but it doesn't matter). The same thing happens on older MPC-HC versions too, but did not try to see how far back.
Ok, the problem occurs with "keep aspect ratio = on". I'm not sure yet why. Could be a bug in madVR or in MPC-HC, I'll investigate. In any case, I'd suggest to set "keep aspect ratio = off", that should fix the problem.

with software decoding madVR IVTC detects a 3:2 pattern. but with disabled deinterlacing it is displaying 23p and only progressive frames.

but why? how could a stream be both at the same?
There are multiple types of telecined content out there. Fields can be encoded one-by-one, or two fields can be encoded in one frame. Those two fields can be from the same original film frame, or from different frames, and those frames can be marked progressive or interlaced. Then there can be flags instructing the decoder to repeat a certain field.

NTSC DVDs, NTSC broadcasts and ATSC broadcasts are known to use any (and sometimes all) of those above variations, and to make things even more interesting, they like to switch back and forth between all those different variants in the middle of the stream sometimes.

madVR's IVTC algorithm has to cope with all these problems. Should madVR change its cadence detection text if the video changes its telecine encoding mode? E.g. if the video has 3:2 separately encoded interlaced fields, madVR should display "3:2", but if the video suddenly changes to 2 encoded frames with a "repeat one field" flag, madVR should change its OSD to "2:2" instead? That would be utterly confusing, and if the OSD changes from "3:2" to "2:2" most users would expect the "cadence break" counter to increase, while in real life madVR is able to stay in the same detected cadence even if the stream changes encoding modes back and forth all the time. Furthermore, if you have a 30i=60i source and madVR detects a 2:2 cadence, wouldn't you expect the final output to be 30p? That's not usually the case.

So practically, madVR treats all telecine encoding modes identically. "3:2" is shown for all modes, regardless of whether each 3:2 sequence actually consists of 5 separately encoded fields, or whether there are only 4 fields (or 2 frames) encoded in the stream with a "repeat one field" flag.

DVD players with interlaced output actually honor these "repeat one field" flags and manually repeat those fields. So the output of a DVD player doesn't differ, regardless of how the stream is encoded. PC decoders are different: They do not manually repeat fields, they just forward the flags. So you can think of madVR's IVTC algorithm in that way that it first interprets the flags and manually copies/duplicates some fields as instructed by the flags, and then applies IVTC to the final stream. And as such the "3:2" text is perfectly right. But madVR of course doesn't do that to save CPU performance. But the end effect is the same.

So I must have brain farted when I said I didn't get why madVR would only detect a 2:2 pattern. If it was progressive, that is all madVR should detect. Retesting the clip over and over, madVR does keep detecting a 3:2 pattern which I have no idea why, but interesting enough, if I add ffdshow in between the MPEG2 decoder and madVR, madVR only detects a 2:2 pattern. This doesn't fix the problem as madVR still messes up at the same spots. Only thing I can think of is this: the 3:2 pattern detection is a display mistake when madVR really is doing a 2:2 pattern, but since a 2:2 pattern is really 2:2:2:2:2:2:2:2:2:2:2:3 pattern, maybe madVR assumes that and messes up every 12 frames?
ffdshow removes the telecine flags, which is not good, obviously.

madVR doesn't assume anything. And a 2:2 pattern is not really 2:2:...:2:2:3, but it's 2:2:2:2....2:2:2:2 "eternally".

Don't try to understand why madVR's IVTC stumbles on this one sample. No algorithm is perfect. I'll analyze the problem when I find some time. But you'll not find some logic to this. It's not a simple logic bug. The problem will come from how madVR compares the separate fields and tries to match them together. This matching algorithm obviously believes that some fields that really do belong together have pixels in them that look like they don't belong together. The problem got nothing to do with flags or cadences or stuff. It's caused by the actual pixels in the fields/frames.

[huhn]

There are multiple types of telecined content out there. Fields can be encoded one-by-one, or two fields can be encoded in one frame. Those two fields can be from the same original film frame, or from different frames, and those frames can be marked progressive or interlaced. Then there can be flags instructing the decoder to repeat a certain field.

NTSC DVDs, NTSC broadcasts and ATSC broadcasts are known to use any (and sometimes all) of those above variations, and to make things even more interesting, they like to switch back and forth between all those different variants in the middle of the stream sometimes.

madVR's IVTC algorithm has to cope with all these problems. Should madVR change its cadence detection text if the video changes its telecine encoding mode? E.g. if the video has 3:2 separately encoded interlaced fields, madVR should display "3:2", but if the video suddenly changes to 2 encoded frames with a "repeat one field" flag, madVR should change its OSD to "2:2" instead? That would be utterly confusing, and if the OSD changes from "3:2" to "2:2" most users would expect the "cadence break" counter to increase, while in real life madVR is able to stay in the same detected cadence even if the stream changes encoding modes back and forth all the time. Furthermore, if you have a 30i=60i source and madVR detects a 2:2 cadence, wouldn't you expect the final output to be 30p? That's not usually the case.

So practically, madVR treats all telecine encoding modes identically. "3:2" is shown for all modes, regardless of whether each 3:2 sequence actually consists of 5 separately encoded fields, or whether there are only 4 fields (or 2 frames) encoded in the stream with a "repeat one field" flag.

DVD players with interlaced output actually honor these "repeat one field" flags and manually repeat those fields. So the output of a DVD player doesn't differ, regardless of how the stream is encoded. PC decoders are different: They do not manually repeat fields, they just forward the flags. So you can think of madVR's IVTC algorithm in that way that it first interprets the flags and manually copies/duplicates some fields as instructed by the flags, and then applies IVTC to the final stream. And as such the "3:2" text is perfectly right. But madVR of course doesn't do that to save CPU performance. But the end effect is the same.

i have no problem with the current behavior if IVTC or at least deinterlacing is used. i just learned where madVR got the 3:2 pattern info from and that this is more like a bug in quicksync that it shows 2:2. and i learned that the stream is accurately 2:2 but with the risk of changing.

but in this case the following is happening:
if a soft telecine DVD is played directly from the ".vob" it is treated and flagged as progressive 29p. same with a MKV remux from makemkv.

so deinterlacing/madVR IVTC isn't used. so what is the point of this?
if it switches to true interlaced the output picture is ruined if it switches to hard telecine 3:2 the output picture is ruined.

if it would be flagged as interlaced than everything can stay as it is even forcing deinterlacing on it and everything is fine as it is. but if it is progressive why not fix the frame interval as long as IVTC or deinterlacing isn't used?

BTW. if deinterlacing is used on such a file than the output is 47p with just repeated frames the deinterlacer never gets the 3:2 pattern. is this a bug bug or working as intended?

[BluesFanUK]

The closest to this would be to use KCP, which packs MPC-HC (or MPC-BE in the Black version) with LAV and MadVR as well as xysubfilter. It has a neat installer, and sets up everything right. It even has an option for you to select "tiers" depending on your rig, so you can try them all (from low to highest) and decide which one works best for you.

The tier profiles need updating because they don't take into account NNEDI3 or the newest enhancements brought in recently by Madshi, but it's a great way to get started if you find the initial setting intimidating. Use them as a starting point and fine-tune and experiment from there.

It's also simple to update madVR if the version included in the latest KCP installer gets a bit old, you just drop the MadVR files into the KCP/MadVR folder and it's up-to-date. Same goes for LAV or MPC-BE. As long as you save your settings beforehand and restore them afterwards (another nice thing about KCP), you'll be fine.

Here is the normal link: http://haruhichan.com/forum/showthread.php?7545-KCP-Kawaii-Codec-Pack

However a newer 1.0.8 build was posted recently and the link hasn't been updated in the front page. You'll find it here: http://haruhichan.com/forum/showthread.php?7545-KCP-Kawaii-Codec-Pack/page296

Another option is to go for jRiver Red October which includes MadVR, but that's not free.

Brilliant, many thanks.

I've been playing around with those settings for the last few days on the highest tier, NNEDI3 is a big no no though.

I have a 980ti and 4K monitor. At that res and with this GPU the TDP is usually in and around 60-80%, I can use NEDDI3 on SD and 720p content, but 1080p 60fps from my OnePlus One produces an obscene number of frame drops.

Presumably it would work a dream on a 1080p monitor though.

Try the settings suggested here: http://forum.kodi.tv/showthread.php?tid=209596&pid=1849097#pid1849097.

Warner, could I be cheeky enough to get an export of the settings you've created and are using please? I tried setting the profiles up last night using your guide to work around my 60fps issue with 1080, but i'm struggling to get my head around the sheer amount of code needed for all the different resolutions/frame rates. KCP has an option to import/export the settings. Think I just need a starting point.

[Aktan]

Oh well, in that case I'm not sure if madVR can work well at all with this kind of source. I mean live TV sources somehow manage to fill the decoder queue. If the analog capture source can't, then there's not much I can do about it. Maybe I can try to optimize this in some far away (!) future build, but for now madVR expects the queues to be nicely filled. And for that the source needs to be able to deliver frames faster than real time. Maybe you could trick the source to do that by intentionally delaying audio by a full second or something like that?

Hmm that's interesting your test case of live TV source doesn't do that. I'll try another analog device then. How would I go about in delaying the audio? Also, what about the idea to delay playback until decoder queue is full? And lastly, why would it work fine for EVR or VMP9 (DXVA deinterlacing that is)?

[retrue]

"Yes, I am using DXVA upscaling, but when I use any other upscaling with this video (using last versions of MPC-HC, XYSubFilter and MadVR) I don't get any image at all. The sound plays fine but the screen is black. With some of the upscaling methods I can see the info when typing Ctr+J, but in most cases I see nothing at all. All the upscaling methods work fine with other videos that I have tried."

To me this sounds like there's some general problem with your PC, either with the software installation or the hardware. Or does this problem only occur with this one specific video file? Can you use Catmull-Rom upscaling for other videos without any problems?


This is happening recently: With some anime videos using Catmull-Rom the video plays well in window mode, but the image is black in fullscreen mode. Sound continues playing and I recover the image again when I switch back to window mode. Maybe I messed something. I don't know. Or maybe it is a bug in MPC-HC or elsewhere. However, with most videos Catmull-Rom works well in window and fullscreen modes. So, maybe the problem is related to subtitles. Srt subs work always well, other formats only sometimes.

Anyway, the problem I had with the upscaling and position of vobsub subtitles and that I reported a couple of weeks ago is anterior to these current problems.
This is my madVR settings.bin file: http://www37.zippyshare.com/v/auRcaug6/file.html
Btw, I am using Windows 8.1 64 bits, with a i7-4770, 16 Gb of RAM and a Nvidia 9800GT graphics card that I cannot change at the moment.

[retrue]

Hi again.

I have been doing checks with old versions of madVR.
The bug I am talking about it is introduced in the version 0.89.7. It is very specific: It happens when the video has vobsub subtitles, XYSubFilter is used as renderer and DXVA2 is used for upscaling.
With different upscalers (Catmull-Rom, Jinc, etc) or different subtitles renderers the subtitles are shown correctly.
Other problem is that recent madVR versions are unstable when running videos with vobsub subtitles. I randomly get black screens instead of video when playing a file. These black screens are most common when switching between window and fullscreen mode. This problem didn't happen with older madVR versions.

[cca]

Does anyone have a problem forcing film mode deinterlacing with the last 2 builds of madvr? For the life of me, I cannot have it select film mode, even if I use the specific key I set for forcing film mode it still selects video mode.

[huhn]

is the source file 10 bit?

[Probe]

Are there any specific requirements to view HDR videos other than using the latest madvr and LAV builds?...l have them both installed and configured ...l can play the HDR demo file (life of pi) but it is way too dark...l can only see the highlights ...the image brightness doesn't change regardless of the nits selected (400 to 10000). l have an gtx 960 and I have madvr configured in FSE D3D11...no presentation glitches just trying to work with the HDR feature...any suggestions?

[dbcooper]

Or if you're using deinterlacing.

:thanks:

Awesome, thanks buddy.

[py930828]

I am using a 3630QM+970M, when I open mpc-hc with madvr, if I am in windoes mode, the queue will go full up to 8, however, if I go into exclusive mode, the render queue will drop down to 1-4-8, the frame will drop about 1 or 2 in one minutes. Also, please trust me that I have tried every passable setting in MADVR. I have been working with this problem for over 60 hours, it doesn't seems like it's working out.

I am sorry I don't know where to find the setting file, if anyone know please tell me. What I can do now is post the debug file, since it produce large file size, what I did is open a file, allow it to run 2-3 seconds, then i set it to exclusive mode for 2-3 seconds, then I close it.

Also, I found that the exclusive mode will go to black screen if I set the "general setting" as DX9 instead of DX11. This could be a problem that some user might have.

Mashi,pPlease reply me if you need any more information. I am looking forward to work with you and solve this problem.

http://1drv.ms/1RyBja4

[zerowalker]

Is it possible to prevent audio from stop playing when seeking, like EVR does.
Meaning the audio insta-seeks and the video will plays, and when the video image is up to point it will start showing as well.

[Asmodian]

Is it possible to prevent audio from stop playing when seeking, like EVR does.
Meaning the audio insta-seeks and the video will plays, and when the video image is up to point it will start showing as well.

Keeping "delay playback start until render queue is full" off and setting "how many video frames shall be presented in advance" to 1 is the best you can do.

[ashlar42]

There is a known bug that I need to work on. It also requires a new XySubFilter patch, which is why I've not fixed it yet. Will be fixed sooner or later, but might take some time...Hi madshi, yesterday I watched a 2.35:1 movie and noticed the subs jumping back inside the video frame after a minute or so... I'm on the latest version of madVR now, but I'm sure it wasn't always like this since the introduction of the new "render on black bars" functionality. Do you know when the bug was introduced? I might consider downgrading if there's a working version without the problem.
Thanks!

[madshi]

but in this case the following is happening:
if a soft telecine DVD is played directly from the ".vob" it is treated and flagged as progressive 29p. same with a MKV remux from makemkv.

so deinterlacing/madVR IVTC isn't used. so what is the point of this?
What is the point of what?

if it switches to true interlaced the output picture is ruined if it switches to hard telecine 3:2 the output picture is ruined.
If who or what switches? Do you mean if you manually switch madVR to true interlaced or hard telecined?

if it would be flagged as interlaced than everything can stay as it is even forcing deinterlacing on it and everything is fine as it is. but if it is progressive why not fix the frame interval as long as IVTC or deinterlacing isn't used?
Your whole post is really confusing to me. I don't really understand at all what you're asking. Here: Are you asking why madVR doesn't fix the frame interval? Or the decoder? Or the splitter? Or the original video encoder?

BTW. if deinterlacing is used on such a file than the output is 47p with just repeated frames the deinterlacer never gets the 3:2 pattern. is this a bug bug or working as intended?
What do you mean with "gets the 3:2 pattern"? And how do you know what the deinterlacer "gets" - since DXVA doesn't tell anybody anything about what it's doing inside. We never know whether it's in film mode or video mode, actually it's usually a pixel-by-pixel decision in most DXVA deinterlacing implementations, I think. And what kind of bug are you talking about? A bug in the GPU DXVA deinterlacing algorithm? Or a bug in madVR?

Hmm that's interesting your test case of live TV source doesn't do that. I'll try another analog device then. How would I go about in delaying the audio?
Some media players have options for delaying audio, e.g. MPC-HC/BE.

Also, what about the idea to delay playback until decoder queue is full?
The bottleneck is not the rendering, so it doesn't matter if we delay until the decoder or renderer queue is full. The bottleneck is "decoding" (or rather sending frames by the source). So any sort of delay doesn't seem to work for this analog source filter you're using.

And lastly, why would it work fine for EVR or VMP9 (DXVA deinterlacing that is)?
EVR/VMR9 simply receive one frame, process it, render it. No queues involved, usually. It's a very simple design, and it works well in this specific situation. madVR is designed to queue frames and render many of them in advance. The whole madVR rendering logic depends on having full queues. If the queues are near empty, all sorts of bad things start to happen. It's a completely different design.

This is happening recently: With some anime videos using Catmull-Rom the video plays well in window mode, but the image is black in fullscreen mode. Sound continues playing and I recover the image again when I switch back to window mode. Maybe I messed something. I don't know. Or maybe it is a bug in MPC-HC or elsewhere. However, with most videos Catmull-Rom works well in window and fullscreen modes. So, maybe the problem is related to subtitles. Srt subs work always well, other formats only sometimes.

Anyway, the problem I had with the upscaling and position of vobsub subtitles and that I reported a couple of weeks ago is anterior to these current problems.
This is my madVR settings.bin file: http://www37.zippyshare.com/v/auRcaug6/file.html
Btw, I am using Windows 8.1 64 bits, with a i7-4770, 16 Gb of RAM and a Nvidia 9800GT graphics card that I cannot change at the moment.

[...]

I have been doing checks with old versions of madVR.
The bug I am talking about it is introduced in the version 0.89.7. It is very specific: It happens when the video has vobsub subtitles, XYSubFilter is used as renderer and DXVA2 is used for upscaling.
With different upscalers (Catmull-Rom, Jinc, etc) or different subtitles renderers the subtitles are shown correctly.
Other problem is that recent madVR versions are unstable when running videos with vobsub subtitles. I randomly get black screens instead of video when playing a file. These black screens are most common when switching between window and fullscreen mode. This problem didn't happen with older madVR versions.
I don't think there is "the bug you are talking about". I think we're talking about 2 different bugs: One bug is windowed vs fullscreen mode, especially when using e.g. Catmull-Rom upscaling. The other bug has to do with vobsub subtitles when using DXVA scaling. The latter bug should be fixed in the next madVR build. Probably the first bug will still be there, though. And it might not be a bug in madVR, but a problem with your GPU driver or hardware. Anyway, it's worth trying the next madVR build, and then checking which problems are still there and which are gone.

Does anyone have a problem forcing film mode deinterlacing with the last 2 builds of madvr? For the life of me, I cannot have it select film mode, even if I use the specific key I set for forcing film mode it still selects video mode.
Are you using native DXVA decoding, maybe? Forced film mode only supports copyback DXVA or software decoding (or CUVID or QuickSync). Also not all pixel formats and bitdepths are supported.

Are there any specific requirements to view HDR videos other than using the latest madvr and LAV builds?...l have them both installed and configured ...l can play the HDR demo file (life of pi) but it is way too dark...l can only see the highlights ...the image brightness doesn't change regardless of the nits selected (400 to 10000). l have an gtx 960 and I have madvr configured in FSE D3D11...no presentation glitches just trying to work with the HDR feature...any suggestions?
There are no specific requirements that I'm aware of. Make sure you use the latest LAV *nightly* build, but even if you don't, HDR files should still look "ok".

Please make a screenshot of the madVR Ctrl+J OSD when playing a HDR video. Don't attach it to this forum, but upload it to some image uploading/sharing host.

I am using a 3630QM+970M, when I open mpc-hc with madvr, if I am in windoes mode, the queue will go full up to 8, however, if I go into exclusive mode, the render queue will drop down to 1-4-8, the frame will drop about 1 or 2 in one minutes. Also, please trust me that I have tried every passable setting in MADVR. I have been working with this problem for over 60 hours, it doesn't seems like it's working out.

I am sorry I don't know where to find the setting file, if anyone know please tell me. What I can do now is post the debug file, since it produce large file size, what I did is open a file, allow it to run 2-3 seconds, then i set it to exclusive mode for 2-3 seconds, then I close it.

Also, I found that the exclusive mode will go to black screen if I set the "general setting" as DX9 instead of DX11. This could be a problem that some user might have.

Mashi,pPlease reply me if you need any more information. I am looking forward to work with you and solve this problem.

http://1drv.ms/1RyBja4
Presenting frames takes one full VSync inveral in fullscreen mode on your PC. I'm not sure why. There's a setting for "max number of prerendered frames" (I don't remember the exact name) in your GPU control panel somewhere. Make sure it's set to application controlled. Generally pretty much everything in your GPU driver's control panel should be set to application controlled, so that madVR is able to choose its own setting instead of being overwritten by the GPU control panel settings.

Hi madshi, yesterday I watched a 2.35:1 movie and noticed the subs jumping back inside the video frame after a minute or so... I'm on the latest version of madVR now, but I'm sure it wasn't always like this since the introduction of the new "render on black bars" functionality. Do you know when the bug was introduced? I might consider downgrading if there's a working version without the problem.
Thanks!
There's a known bug in the way XySubFilter and madVR communicate. This will be fixed in a future XySubFilter + madVR version.

[madshi]

madVR v0.89.19 released

http://madshi.net/madVR.zip

* added support for display peak luminance values of 265, 180 and 120 nits
* HDR: gamut mapping is now always on, even if calibration control is disabled
* fixed: DVD/Blu-Ray subs were positioned incorrectly when using DXVA scaling
* fixed: DVD aspect ratio wasn't always shown correctly when using MPC-HC
* fixed: gamma processing and brightness control didn't work for HDR content
* fixed: DXVA decoding + deint -> refresh rate sometimes didn't switch
* repeated frames are displayed (again) when smooth motion FRC is enabled

[Aktan]

Some media players have options for delaying audio, e.g. MPC-HC/BE.

Okay, I'll try that, thanks!

The bottleneck is not the rendering, so it doesn't matter if we delay until the decoder or renderer queue is full. The bottleneck is "decoding" (or rather sending frames by the source). So any sort of delay doesn't seem to work for this analog source filter you're using.

What made me think this may help is this. With pure interlace source, using DXVA deinterlacing, if the queues are under 7, I see DXVA change how it works and only see a bobing deinterlace (a lot of shimmering). If the queues are 7 or higher, DXVA deinterlace works completely fine and looks great. I thought this happens because DXVA needed more frames before it could process. So what I thought was that since the decoder queue is near zero and with it the render queue near zero, DXVA is messing up, but if there is a delay until decoding queue is full, it would consistently be around full (since it is capturing realtime) but the queue would have enough frames for DXVA to work it's magic.

EVR/VMR9 simply receive one frame, process it, render it. No queues involved, usually. It's a very simple design, and it works well in this specific situation. madVR is designed to queue frames and render many of them in advance. The whole madVR rendering logic depends on having full queues. If the queues are near empty, all sorts of bad things start to happen. It's a completely different design.

I forgot where I saw this, this was years ago, but I'm pretty sure any video render that works with DXVA does have a small queue of maybe 1-2 frames. In order to do motion estimation on temporal side, it needs more than 1 frame. I remember the option to set how many frames to queue and if it was set to zero, DXVA wouldn't work right.

[Budtz]

Does the Nits-setting matter with non-HDR content?

I can't find any info on it for my D6505 samsung TV

[omarank]

added support for display peak luminance values of 265, 180 and 120 nits

Thanks! I very much like the look of 265 nits setting. The shadow details are wonderful at this setting.

Incidentally, have you made any improvements/ changes for the regular SDR content too? I seem to feel a subtle PQ improvement. Can’t say if it’s a placebo.

[madshi]

I forgot where I saw this, this was years ago, but I'm pretty sure any video render that works with DXVA does have a small queue of maybe 1-2 frames. In order to do motion estimation on temporal side, it needs more than 1 frame. I remember the option to set how many frames to queue and if it was set to zero, DXVA wouldn't work right.
DXVA GPU processors have the option to ask for past frame references, but IIRC newer AMD and NVidia drivers don't really ask for that, they seem to store/remember past frames themselves somehow.

Does the Nits-setting matter with non-HDR content?

I can't find any info on it for my D6505 samsung TV
At this time the Nits setting only applies to HDR content. Just try different settings to find one which looks good to your eyes.

Thanks! I very much like the look of 265 nits setting. The shadow details are wonderful at this setting.
Glad to hear that!

Incidentally, have you made any improvements/ changes for the regular SDR content too? I seem to feel a subtle PQ improvement. Can’t say if it’s a placebo.
I haven't changed anything SDR PQ related. At least not intentionally... :D

[zerowalker]

Keeping "delay playback start until render queue is full" off and setting "how many video frames shall be presented in advance" to 1 is the best you can do.

Ah okay, already tried that, guess i am too greedy, thanks:)

[Aktan]

DXVA GPU processors have the option to ask for past frame references, but IIRC newer AMD and NVidia drivers don't really ask for that, they seem to store/remember past frames themselves somehow.

Ah okay. I guess what I mention about madVR needs a queue of 7 or more to have DXVA work correctly doesn't help my problem?

[madshi]

Ah okay. I guess what I mention about madVR needs a queue of 7 or more to have DXVA work correctly doesn't help my problem?
Not really. madVR can work ok with just e.g. 4 frames, too. But the frames must be coming in reliably from the source filter. If the decoder queue gets down to 0 or 1 frames once in a while, bad things will happen.

[aufkrawall]

Thanks for the new build. Could the next please include the latest Adaptive Sharpen version?

[madshi]

You like it better than "sharpen edges"?

[retrue]

I don't think there is "the bug you are talking about". I think we're talking about 2 different bugs: One bug is windowed vs fullscreen mode, especially when using e.g. Catmull-Rom upscaling. The other bug has to do with vobsub subtitles when using DXVA scaling. The latter bug should be fixed in the next madVR build. Probably the first bug will still be there, though. And it might not be a bug in madVR, but a problem with your GPU driver or hardware. Anyway, it's worth trying the next madVR build, and then checking which problems are still there and which are gone.


Thank you very much for fixing the vobsub bug when using DXVA scaling.
I am using the last drivers for my graphics card.
About the frequent black screens problem, I don't think it is a hardware or driver problem on my side because I don't have this problem with older madVR versions. I think until madVR version 0.89.6 it worked well. In the last madVR versions I have got a lot of black screens. Let's hope this new version is more stable for me.
And thanks again.

[huhn]

What is the point of what?
I don't know why a soft telecine DVD remux oder vob is treated as progressive but played back at 29p by default.

that looks pointless to me.
If who or what switches? Do you mean if you manually switch madVR to true interlaced or hard telecined?
a soft telecine source is flagged progressive that's why deinterlacing isn't used. so we just get 30 FPS with 23p in it. so even if the same stream is switching between telecine modes the playback is not correct because it will not be deinterlaced/IVTC'd.

Your whole post is really confusing to me. I don't really understand at all what you're asking. Here: Are you asking why madVR doesn't fix the frame interval? Or the decoder? Or the splitter? Or the original video encoder?

I don't know what part should change his behaviour.
but playing a 23p stream as 29p without deinterlancing is just wrong.
madVR sees or at least can see the repeat flags so it should use deinterlancing, IVTC or just change the refreshrate/frame interval to 23p if doesn't change to interlace mode.

i don't know if this should be done by the source filter, decoder or madVR.

lavfilter software is outputting a 23p stream with repeat flags and I don't think there is anything wrong with it but it still says the source is progressive.
What do you mean with "gets the 3:2 pattern"? And how do you know what the deinterlacer "gets" - since DXVA doesn't tell anybody anything about what it's doing inside. We never know whether it's in film mode or video mode, actually it's usually a pixel-by-pixel decision in most DXVA deinterlacing implementations, I think. And what kind of bug are you talking about? A bug in the GPU DXVA deinterlacing algorithm? Or a bug in madVR?

I just counted the frames by using control + arrow key.
with a soft telecine source, every frame is just repeated unlike 3:2 hard telecine.

I don't know if the deinterlancer knows it is 3:2 soft telecine or it just gets 2:2 from madVR, no clue. but the output is not 59p.

[aufkrawall]

You like it better than "sharpen edges"?
Nope, but I like it a lot more than Lumasharpen or crispen edges.
Sharpen edges is still a keeper for me, however it's still more expensive. There was at least the report of one user who couldn't use sharpen edges with his GPU, while AS was fast enough.

I'm aware that you don't necessarily share the opinion that everything must be open, which is totally fine as I agree that intellectual property should be valued. But is there a chance that you explain at some point how sharpen edges works? It seems quite unique to me and could be really helpful for e.g. ReShade.
Feel free to say "no". ;)
Adaptive Sharpen is great for games too, as temporal AA seems to blur in gamma light and a slight birghtness increase due to sharpen might be even more correct.

[KoD]

madVR v0.89.19 released

http://madshi.net/madVR.zip

* added support for display peak luminance values of 265, 180 and 120 nits
* HDR: gamut mapping is now always on, even if calibration control is disabled
* fixed: DVD/Blu-Ray subs were positioned incorrectly when using DXVA scaling
* fixed: DVD aspect ratio wasn't always shown correctly when using MPC-HC
* fixed: gamma processing and brightness control didn't work for HDR content
* fixed: DXVA decoding + deint -> refresh rate sometimes didn't switch
* repeated frames are displayed (again) when smooth motion FRC is enabled
Thank you, madshi! I can confirm the issues on the sample files I provided are all fixed.

[py930828]

What is the point of what?


If who or what switches? Do you mean if you manually switch madVR to true interlaced or hard telecined?


Your whole post is really confusing to me. I don't really understand at all what you're asking. Here: Are you asking why madVR doesn't fix the frame interval? Or the decoder? Or the splitter? Or the original video encoder?


What do you mean with "gets the 3:2 pattern"? And how do you know what the deinterlacer "gets" - since DXVA doesn't tell anybody anything about what it's doing inside. We never know whether it's in film mode or video mode, actually it's usually a pixel-by-pixel decision in most DXVA deinterlacing implementations, I think. And what kind of bug are you talking about? A bug in the GPU DXVA deinterlacing algorithm? Or a bug in madVR?


Some media players have options for delaying audio, e.g. MPC-HC/BE.


The bottleneck is not the rendering, so it doesn't matter if we delay until the decoder or renderer queue is full. The bottleneck is "decoding" (or rather sending frames by the source). So any sort of delay doesn't seem to work for this analog source filter you're using.


EVR/VMR9 simply receive one frame, process it, render it. No queues involved, usually. It's a very simple design, and it works well in this specific situation. madVR is designed to queue frames and render many of them in advance. The whole madVR rendering logic depends on having full queues. If the queues are near empty, all sorts of bad things start to happen. It's a completely different design.


I don't think there is "the bug you are talking about". I think we're talking about 2 different bugs: One bug is windowed vs fullscreen mode, especially when using e.g. Catmull-Rom upscaling. The other bug has to do with vobsub subtitles when using DXVA scaling. The latter bug should be fixed in the next madVR build. Probably the first bug will still be there, though. And it might not be a bug in madVR, but a problem with your GPU driver or hardware. Anyway, it's worth trying the next madVR build, and then checking which problems are still there and which are gone.


Are you using native DXVA decoding, maybe? Forced film mode only supports copyback DXVA or software decoding (or CUVID or QuickSync). Also not all pixel formats and bitdepths are supported.


There are no specific requirements that I'm aware of. Make sure you use the latest LAV *nightly* build, but even if you don't, HDR files should still look "ok".

Please make a screenshot of the madVR Ctrl+J OSD when playing a HDR video. Don't attach it to this forum, but upload it to some image uploading/sharing host.


Presenting frames takes one full VSync inveral in fullscreen mode on your PC. I'm not sure why. There's a setting for "max number of prerendered frames" (I don't remember the exact name) in your GPU control panel somewhere. Make sure it's set to application controlled. Generally pretty much everything in your GPU driver's control panel should be set to application controlled, so that madVR is able to choose its own setting instead of being overwritten by the GPU control panel settings.


There's a known bug in the way XySubFilter and madVR communicate. This will be fixed in a future XySubFilter + madVR version.


Hi Madshi, thank you for replying me, I set the nvidia control as everything is set by the 3D application. It still doesn't work out. However, should I add MADVR into NVIDIA control panel? Since so far, I only set MPC-HC run by NVIDIA graphics card but not MADVR, I don't know what application I should choose to put into the control panel.

Also, I have the other question. I like to watch high quality source movie, and I don't watch anime. Most of my movie sources are blueray, so what recommend setting do you suggest me to set on? Currently I have

chroma upscaling:super-xbr sharpness:100 AR+SuperRes filter
image downscaling:Catmull-Rom AR+linear light
image doubling:none
image upscaling:Jinc AR+LINEAR LIGHT
upscaling refinement: SuperRes

I simply copy what some other user's setting since that these setting really confused me. It would be really great if you can give me some of your valuable suggestion.
Thank you for your time.

[Barnahadnagy]

I for one like Adaptive Sharpen a lot, tho pretty much only in image processing and not upscaling refinement (and only when no upscaling is done). There are sources where Sharpen Edges is better, but overall I'm usually rolling with AS + Thin Edges (this thing is godly).

[har3inger]

Hi Madshi, thank you for replying me, I set the nvidia control as everything is set by the 3D application. It still doesn't work out. However, should I add MADVR into NVIDIA control panel? Since so far, I only set MPC-HC run by NVIDIA graphics card but not MADVR, I don't know what application I should choose to put into the control panel.

Also, I have the other question. I like to watch high quality source movie, and I don't watch anime. Most of my movie sources are blueray, so what recommend setting do you suggest me to set on? Currently I have

chroma upscaling:super-xbr sharpness:100 AR+SuperRes filter
image downscaling:Catmull-Rom AR+linear light
image doubling:none
image upscaling:Jinc AR+LINEAR LIGHT
upscaling refinement: SuperRes

I simply copy what some other user's setting since that these setting really confused me. It would be really great if you can give me some of your valuable suggestion.
Thank you for your time.

Don't use linear light for upscaling. Jinc AR is fine. If you're watching 1080p videos on 1080p TV, this setting isn't ever used anyways. The same goes for upscaling refinement SuperRes. If not, use the image doubling options to use nnedi3 (64 neurons is best, 32 is a compromise but still good, 16 isn't great unless for anime) or superxbr 100.

Chroma upscaling SuperRes probably shouldn't be used for now, as it has some issues with dark lines at edges of colored areas.

Personally I prefer SXBR 125 without AR checked for chroma. It rings a lot less than with AR checked, but still looks better than jinc AR. Whether you like this more depends on what you like.

If you have performance to spare, you can use nnedi3 for chroma upscaling. It's generally pretty wasteful, and one of the last things you should consider turning on when you have performance to spare.


@madshi: the OSD font size (and maybe the font itself) changed a lot with the latest version. Is this intentional?

[Georgel]

@Madshi

Please, include an option to disable dithering completely.

Every single type of dithering is critically affecting all anime materials.

I can understand how it works wonders with movie and real life videos, but with anime it is just adding noise over a surface that was never meant to have noise (having different colored pixels on a smooth skin surface, glitter in areas where there should be 0 darkness)

I know that it would be hard to modify dithering, but I was watching fate Kaleid Liner Prisma Illya and there was bad noise everywhere, the rip is q16 off BD and is 1920X1080 done very good.

Or am I missing something, and dithering with error diffusion 2 should be the lowest noise of all?

I am still learning, but I know that for the moment I get lots of noise, even with error diffusion 2 with only luma and with or without changing dither every frame.

[har3inger]

@Madshi

Please, include an option to disable dithering completely.

Every single type of dithering is critically affecting all anime materials.

I can understand how it works wonders with movie and real life videos, but with anime it is just adding noise over a surface that was never meant to have noise (having different colored pixels on a smooth skin surface, glitter in areas where there should be 0 darkness)

I know that it would be hard to modify dithering, but I was watching fate Kaleid Liner Prisma Illya and there was bad noise everywhere, the rip is q16 off BD and is 1920X1080 done very good.

Or am I missing something, and dithering with error diffusion 2 should be the lowest noise of all?

I am still learning, but I know that for the moment I get lots of noise, even with error diffusion 2 with only luma and with or without changing dither every frame.

The source probably has a lot of source dither/grain/noise. This may be the producer's intention. Unless your screen bit depth is set really low, any of madvr's dithering algorithms shouldn't add much noise at all.

You can already turn off dither in MadVR. Rendering > Dithering > None. Use that, and compare to see if there's more or less noise. However, using no dither is always a terrible idea. You get banding and posterization artifacts.

[dbcooper]

Hi Madshi, if you choose "use half frame rate for deinterlacing", the frame time on the OSD does not change. This is with DXVA deinterlacing.

[Aktan]

Not really. madVR can work ok with just e.g. 4 frames, too. But the frames must be coming in reliably from the source filter. If the decoder queue gets down to 0 or 1 frames once in a while, bad things will happen.

Which is why I think if there was an option to delay playback until decoder queue is full, then there be no problem. This is a capture source. If this capture source is to be reliable to capture at full framerate all the time, the computer must be fast enough and compress it fast enough, else there be drop frames. I can trust my computer can capture fine and that once the decoder queue is full, would stay full even if it can't deliver frames faster than realtime.

Edit: I just realized, why isn't the render queue delayed until full? When I first start playback, it isn't like the render queue is full then slowly goes down to 0-2. It's 0-2 from the start.

[huhn]

@Madshi

Please, include an option to disable dithering completely.

Every single type of dithering is critically affecting all anime materials.

I can understand how it works wonders with movie and real life videos, but with anime it is just adding noise over a surface that was never meant to have noise (having different colored pixels on a smooth skin surface, glitter in areas where there should be 0 darkness)

I know that it would be hard to modify dithering, but I was watching fate Kaleid Liner Prisma Illya and there was bad noise everywhere, the rip is q16 off BD and is 1920X1080 done very good.

Or am I missing something, and dithering with error diffusion 2 should be the lowest noise of all?

I am still learning, but I know that for the moment I get lots of noise, even with error diffusion 2 with only luma and with or without changing dither every frame.

you can disable dithering by selecting "none".

but dithering is way more important for anime than real life source. dithering prevents banding and banding is the biggest issue in anime BDs.
even mastering should use dithering and adding noise in animes isn't that rare.

[Asmodian]

@Madshi

Please, include an option to disable dithering completely.

Every single type of dithering is critically affecting all anime materials.

I can understand how it works wonders with movie and real life videos, but with anime it is just adding noise over a surface that was never meant to have noise (having different colored pixels on a smooth skin surface, glitter in areas where there should be 0 darkness)

I know that it would be hard to modify dithering, but I was watching fate Kaleid Liner Prisma Illya and there was bad noise everywhere, the rip is q16 off BD and is 1920X1080 done very good.

Or am I missing something, and dithering with error diffusion 2 should be the lowest noise of all?

I am still learning, but I know that for the moment I get lots of noise, even with error diffusion 2 with only luma and with or without changing dither every frame.

Ordered Dithering is the lowest noise option and it is much better for Anime than rounding (None). Use Ordered Dithering, turn off "change dither for every frame", and leave on "use colored noise" for the lowest visible noise configuration. Personally I like both options off, colored noise is worse than luma noise, but the difference is subtle.

I almost always watch Anime myself.

I would be very surprised if you could see the noise from 8-bit ordered dithering with "change dither for every frame" disabled. I cannot. The noise you observed in Kaleid Liner Prisma Illya was probably not due to dithering, you don't see it in every source do you?

Please do not use "None", it makes me sad when people disable dithering. :(

[py930828]

Don't use linear light for upscaling. Jinc AR is fine. If you're watching 1080p videos on 1080p TV, this setting isn't ever used anyways. The same goes for upscaling refinement SuperRes. If not, use the image doubling options to use nnedi3 (64 neurons is best, 32 is a compromise but still good, 16 isn't great unless for anime) or superxbr 100.

Chroma upscaling SuperRes probably shouldn't be used for now, as it has some issues with dark lines at edges of colored areas.

Personally I prefer SXBR 125 without AR checked for chroma. It rings a lot less than with AR checked, but still looks better than jinc AR. Whether you like this more depends on what you like.

If you have performance to spare, you can use nnedi3 for chroma upscaling. It's generally pretty wasteful, and one of the last things you should consider turning on when you have performance to spare.


@madshi: the OSD font size (and maybe the font itself) changed a lot with the latest version. Is this intentional?

Thank you for your reply, it was really kind to give me suggestions. Could you and Madshi please solve my major problem which is about the frame drop? @Madshi

These two pictures, when I was in windows mode, the render and present queue is full, but if I am in fullscreen, the queue drop to 2, and the present time is really really long. My MPC-HC and MADVR are all in newest version. This problem really annoying me,and I couldn't enjoy watch movies since the frame drop significantly randomly. I would be really appreciated if I could get rid of this problem.

If anyone know how to solve this problem, please, please please help me out!!

@Madshi


@Madshi, I found the problem!!! The reason why the render queue and the present queue drop is because of DX11, if I set it to DX9 in "general setting", the queue will get back to full, and the presenting time goes back from 16ms to 2ms(in full screen case). However, there are two problems that I have found so far:

1. In DX9 mode, I can't use the exclusive mode since once I able exclusive mode, the full screen is just going to be black screen.

2. Instead of the increasing of dropped frame in DX11 mode, DX9 mode increases presentation glitches a lot, one completely movie might have 200 more presentation glitches.

Again, just a reminder, I am using a laptop with 3630QM+970M, do you think this can be a problem?

This might be a problem that I hope can be fixed in the next version.
Also, could you please tell me what is the benefit of setting it to DX11? Does the graphics looks better or it's mean to be more efficient?

There are some screen shots that might give you some information
Madshi, hope this can be helpful and not boring you.

[Nullack]

Madshi can you please accommodate GUI experiences with UHD resolutions and high dpi fonts? For example, the CTRL J stats render is unviewable at in high DPI (e.g. 300%) and UHD 4K resolutions. At the moment Im having to take a screenshot then zoom in to see the render stats.

[6233638]

Madshi can you please accommodate GUI experiences with UHD resolutions and high dpi fonts? For example, the CTRL J stats render is unviewable at in high DPI (e.g. 300%) and UHD 4K resolutions. At the moment Im having to take a screenshot then zoom in to see the render stats.I actually thought he had done that in this version.
I was just about to write up a post asking about the GUI changes because I find them difficult to read on my 1080p screen.
With this build I'm now seeing a big proportionally spaced and anti-aliased font (http://abload.de/img/new-osd-full20oi4.png) (with bad keming (https://en.wikipedia.org/wiki/Kerning)) vs the old monospaced bitmap font (http://abload.de/img/old-osd-fullv6ohk.png).
It's worse in a window because the new stats (http://abload.de/img/new-stats-window6aoph.png) now scale with the window size and get a lot smaller than the old ones (http://abload.de/img/old-stats-windowxboxi.png).

[SweetLow]

I've no idea if that's possible. And if it worked I've no idea if that would screw up things like calibration and stuff. Might be worth trying at some point, although I don't know if the OS/GPU driver even gives me access to DDC/DI. But this is not going to happen any time soon.

It's possible (and i use it more then decade already - to increase monitor luminance on the fly to view video and pictures :) ), but, AFAIK, there is no standard user mode API to access this feature under Win platform.

>And if it worked I've no idea if that would screw up things like calibration and stuff.
Are the change of backlight screw up calibration?

>But this is not going to happen any time soon.
And complete realization is not need (in first) - minimum adaptation is
1. Set max and min luminances of monitor in config screen instead of only one value. (And third, default luminance value - to set it on exit).
2. Write code to call external executables and pass there stream luminance(or min/max monitor value if stream luminance out of monitor range), as parameter in execution point where luminance is changed - and assume this value is current monitor luminance ;)

[enphenate]

I have a Samsung UN50JU6500 UHD TV and every time MadVR changes my refresh rate to 23Hz, my TV gets kicked off of 4:4:4 at 1080p. At 1080p 60hz I can do 4:4:4 or full RGB without a problem, but when it changes to any different refresh rate, my TV no longer stays in PC mode (4:4:4). Its immediately noticeable because when it fails to pass 4:4:4 all of the image customization (RGB / White Balance) become available.

Anyone have an idea why? Does my TV only support 4:4:4 at 60hz and no other refresh rates or could this be a setting in MadVR?

[SweetLow]

Anyone have an idea why? Does my TV only support 4:4:4 at 60hz and no other refresh rates or could this be a setting in MadVR?
It's Samsung "feature" :(