madVR - high quality video renderer (GPU assisted)

[6233638]

I have a few samples where a maxAngle of 0.10 seems a tad too low. I think 0.15 might be necessary? Maybe even 0.20? But maybe a value that high would start destroying detail again?Unfortunately I spent quite a lot of time doing testing, only to have you bring out a new build shortly after. (not that I am complaining) In my brief testing 0.10 seemed to be producing similar results to the ones I had before, but maybe it is too conservative.

We have yet to approach medium/high settings, which is where I was planning on increasing it. I'll have another look over everything again.

Anyway, what is your opinion about AngleBoost now, with the new build? How useful is it?

And it seems you've set LocalContrast off now, too? So maybe I can remove it, after all?It seems like a combination of AngleBoost and nonGradientPenalty can replace Local Contrast now, and do so with less artifacting.

I'm hesitant to say it should be removed yet but I suspect it won't be necessary any more.

Rerendering all the time would mean that GPU usage would jump up during paused state.Oh, I just meant for the test builds. I seem to recall you making that change years ago when people complained about high usage when video was paused.

[Werewolfy]

If you want absolutely no loss details loss, I found this :


avgDif maxDif grad Penalty LC Boost maxAngle
0.5 2.0 1.1 1.0 off 1.9 0.16


I haven't seen one single detail that was blurred with this setting. It's mainly for informative purpose, I don't know if I would really recommend it for a low setting because sometimes the debanding effect is quite efficient (thanks to angleBoost) but unfortunately sometimes it does very little.

[madshi]

Unfortunately I spent quite a lot of time doing testing, only to have you bring out a new build shortly after.
Yes, I know that means some of your testing has been "wasted". I'm not happy about it at all. But then I think the improvements of the AngleBoost feature was worth it. FWIW, I don't plan any further algorithm changes at this moment.

Oh, I just meant for the test builds.
I've tried but didn't manage to make it work without digging too deep into the rendering pipeline.

If you want absolutely no loss details loss, I found this
Maybe that's a bit too careful? I'm not sure. Opinions?

---------------------

So your current recommendations with the latest test build seem to be:

avgDif maxDif grad Penalty LC Boost maxAngle
6233638: 0.6 2.2 1.4 1.2 off 2.2 0.10
Werewolfy: 0.6 2.0 1.3 1.2 off 1.9 0.16
bacondither: 0.7 2.0 1.4 1.2 off 1.5 0.20
I suppose all of these were suggestions for the "low" preset, correct? Can we try to find common ground here? If I try to build some kind of average of your 3 suggestions I'd say:

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 2.1 1.4 1.2 off 1.9 0.16
Can you double check whether this would make a good "low" preset? The "low" preset should be conservative enough so that we can recommend madVR users to always enable it for all content without having to worry about losing (much) image detail.

[bacondither]

@madshi

Could the "low" "med" "high" debanding settings be presets instead of being set in stone, so everyone could be happy and we would not have to live with a compromise. :sly:

[6233638]

I've tried but didn't manage to make it work without digging too deep into the rendering pipeline.OK, thanks anyway.

Maybe that's a bit too careful? I'm not sure. Opinions?I'm not sure it has effective enough debanding. That said, it's interesting how much adding angle boost has changed things.

I suppose all of these were suggestions for the "low" preset, correct? Can we try to find common ground here? If I try to build some kind of average of your 3 suggestions I'd say:

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 2.1 1.4 1.2 off 1.9 0.16
Can you double check whether this would make a good "low" preset? The "low" preset should be conservative enough so that we can recommend madVR users to always enable it for all content without having to worry about losing (much) image detail.With anything over 0.10 for max angle, I'm seeing texture in dark hair start to blend together.
If you're wanting low to be a setting that people can leave on all the time, I'm not sure it can be that high.

But if you really think 0.10 is too low, I'd suggest 0.18.
If you're willing to sacrifice some detail, I think it does a better job of debanding than 0.16

I don't have a problem with the maxDif/Boost setting there, but I do find that 2.2 works better for them, without hurting detail. (at least when max angle is 0.10)

There are so many variables now, it's difficult to compare everything though. Before the Angle Boost setting it was a lot easier as you could isolate each varible.
Maybe we don't have things "perfect" yet, but I don't know that it's worth spending much more time on the Low preset.

It's impressive how far we have come in such a short time. We have much smoother debanding and better detail retention:

Source (http://www.abload.de/img/originalo4rim.png)
Old Settings (http://www.abload.de/img/oldxapvx.png)
New Settings (0.10) (http://www.abload.de/img/newswo29.png)
New Settings (0.18) (http://www.abload.de/img/highfgqi5.png)



Could the "low" "med" "high" debanding settings be presets instead of being set in stone, so everyone could be happy and we would not have to live with a compromise. :sly:I assumed we would return to "low/med/high/custom" eventually, but maybe that's a better way of handling it.

Have defaults for low/med/high, but allow users to customize them if they feel it necessary.

[leeperry]

@madshi: I'm not sure if that would apply as a feature request but it would be great if custom gamma came with +1/-1 buttons so we could reach any 0.0x increment, because I've got yet another new TV here and 2.25 is sometimes too dark and 2.20 too clear :o

It's about time I'd give BT.1886 a whirl I guess :devil:

I suspect that his instrumentation cannot handle the refresh rate, but it's always possible. I haven't looked into that aspect because the color quality is really bad on those displays anyway. The point is that you get the absolute best latency and response times - which is what matters for competitive gaming - at the cost of all else.
His meter can easily handle 120Hz CRT and he claims contrast to look extremely poor as compared to 1K:1 60Hz so this would add up.

the point is that it's dynamic so you no longer have refresh rates, you just run the video at the native framerate. Depending on how precise this is, you could potentially do without ReClock/VideoClock and sync up the video to the audio clock.
Very few movies come in 24.000fps untouched flavor so Reclock is pretty much always necessary anyway and nvidia is not willing to share the golden goose so you can rest assured that they'll make licensees pay through the nose for this new display controller. Video and audio clocks will still be separated so I don't really see how Reclock could be put out of the loop, well except if you go HDMI audio as this would enslave both clocks together FWIR.

Also, I think I prefer having mVR doing the frames repeating@120Hz than some sloppy controller with dubious jitter specs...oh well, 120Hz OLED will make me plenty happy :)

[iSunrise]

It's impressive how far we have come in such a short time. We have much smoother debanding and better detail retention:

Source (http://www.abload.de/img/originalo4rim.png)
Old Settings (http://www.abload.de/img/oldxapvx.png)
New Settings (0.10) (http://www.abload.de/img/newswo29.png)
New Settings (0.18) (http://www.abload.de/img/highfgqi5.png)

Just...WOW.

Comparing Source and New Settings (0.18) looks like a night and day difference. It almost completely eliminates the blocks. Looks like a pretty powerful deblocking filter with debanding now. Wish I had something like this some years ago. When I still was using FFDShow, the included deblock filters all had serious side-effects (too much blur being one of them).

If you can improve that even more (without taking away details) this is kind of a no-brainer to enable at default. Was a bit sceptic at first (I usually don´t like such things enabled by default), but if those screens are any indication, no worries here.

Might be interesting to judge them on a bigger, high-quality screen (at least 65"), only on a 24" here.

[Werewolfy]

Maybe that's a bit too careful? I'm not sure. Opinions?

I agree ;) I just wanted to report it but now even with this setting I can find details loss so let's forget it.

@madshi

Could the "low" "med" "high" debanding settings be presets instead of being set in stone, so everyone could be happy and we would not have to live with a compromise. :sly:

I know it's not the purpose of madshi but it's a very good idea!


With anything over 0.10 for max angle, I'm seeing texture in dark hair start to blend together.
If you're wanting low to be a setting that people can leave on all the time, I'm not sure it can be that high.

But if you really think 0.10 is too low, I'd suggest 0.18.
If you're willing to sacrifice some detail, I think it does a better job of debanding than 0.16

I don't have a problem with the maxDif/Boost setting there, but I do find that 2.2 works better for them, without hurting detail. (at least when max angle is 0.10)

There are so many variables now, it's difficult to compare everything though. Before the Angle Boost setting it was a lot easier as you could isolate each varible.
Maybe we don't have things "perfect" yet, but I don't know that it's worth spending much more time on the Low preset.

It's impressive how far we have come in such a short time. We have much smoother debanding and better detail retention:

Source (http://www.abload.de/img/originalo4rim.png)
Old Settings (http://www.abload.de/img/oldxapvx.png)
New Settings (0.10) (http://www.abload.de/img/newswo29.png)
New Settings (0.18) (http://www.abload.de/img/highfgqi5.png)


For maxDif, I don't have a problem with a value of 2.2 but I never found an improvement over 2.0. If you find 2.2 is better on some videos, we can use it, maxDif dosen't seem to blur details.

For angleBoost I strongly suggest to limit ourselves to 1.9. From 2.0, it blurs dark sharp lines. I saw it only on a test pattern and no real life content I admit but for a low preset, I think that we should not "take the risk".

maxAngle on 0.10 is not very effective sometimes but it's true that for a low preset we aim almost no details loss so maybe we should stick to 0.10 even if a higher value is very tempting... Can you post a video that shows the problem with higher values of maxAngle please? Even with a lower value for angleBoost or Gradient, it starts to blur from 0.11?

I think we're very close to find a "perfect" low preset ;)

[Ver Greeneyes]

For angleBoost I strongly suggest to limit ourselves to 1.9. From 2.0, it blurs dark sharp lines. I saw it only on a test pattern and no real life content I admit but for a low preset, I think that we should not "take the risk".
I'm not going to argue with your testing since I haven't taken the time to do any, but I wanted to point out that it doesn't really make sense for 1.9 vs 2.0 to be some magic threshold - after all it's a boost, so the strength of its effect depends on your other settings. It's more likely that for the settings you've chosen, 2.0 happens to be too strong - if 6233638's other settings are lower, 2.0 might be okay for him.

[madshi]

Comparing Source and New Settings (0.18) looks like a night and day difference. It almost completely eliminates the blocks. Looks like a pretty powerful deblocking filter with debanding now. Wish I had something like this some years ago. When I still was using FFDShow, the included deblock filters all had serious side-effects (too much blur being one of them).
FWIW, those images are brightness/contrast boosted to more clearly show the banding removal, I think. The debanding algorithm doesn't work too well for blocking artifacts. Not as good as these test images seem to suggest. The debanding algorithm is really only meant to remove banding artifacts - not block artifacts.

Could the "low" "med" "high" debanding settings be presets instead of being set in stone, so everyone could be happy and we would not have to live with a compromise. :sly:
I assumed we would return to "low/med/high/custom" eventually, but maybe that's a better way of handling it.
I wasn't really planning to offer any "custom" mode. But if you guys insist maybe you can pursuade me...

With anything over 0.10 for max angle, I'm seeing texture in dark hair start to blend together.
If you're wanting low to be a setting that people can leave on all the time, I'm not sure it can be that high.

But if you really think 0.10 is too low, I'd suggest 0.18.
If you're willing to sacrifice some detail, I think it does a better job of debanding than 0.16

I don't have a problem with the maxDif/Boost setting there, but I do find that 2.2 works better for them, without hurting detail. (at least when max angle is 0.10)
The problem with the new AngleBoost feature is that it's highly dependent on the exact source. With some sources a small maxAngle works well, with other sources it doesn't. With some sources a high boost value doesn't harm, with others it may harm. That makes it a bit difficult to choose "good" values. The other thresholds are a bit more predictable, it seems to me.

Generally, with the "low" setting I'd prefer to concentrate on preserving as much detail as possible. So if you say that you see detail loss with maxAngle 0.16, but not with maxAngle 0.10, then I tend to pick 0.10 for the "low" setting. Personally, I have one example where a boost of 2.2 hurts, so I'd prefer to stick to 1.9 for the "low" setting.

So my current favorite for LOW:

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 2.1 1.4 1.2 off 1.9 0.10
Can everybody live with this?

-------------------------

Might be time to look into "medium" and "high" settings. For medium I'm wondering whether we should increase all thresholds by a similar factor. Or whether we should maybe only increase thresholds slightly and use a wider "maxAngle" for the AngleBoost instead?

So any suggestions for "medium" and "high" anybody?

[madshi]

maxDif dosen't seem to blur details.
Does everybody agree with this? If so, we could simply skip this check completely.

[Werewolfy]

I'm not going to argue with your testing since I haven't taken the time to do any, but I wanted to point out that it doesn't really make sense for 1.9 vs 2.0 to be some magic threshold - after all it's a boost, so the strength of its effect depends on your other settings. It's more likely that for the settings you've chosen, 2.0 happens to be too strong - if 6233638's other settings are lower, 2.0 might be okay for him.
Agreed, I meant 1.9 is the limit with the other settings we use and posted ealirer for a low preset. 6233638's settings are not very different from mine and it happens also with their settings.


So my current favorite for LOW:

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 2.1 1.4 1.2 off 1.9 0.10
Can everybody live with this?


If maxAngle really hurts from 0.11, this preset works fine for me ;)

[6233638]

FWIW, those images are brightness/contrast boosted to more clearly show the banding removal, I think. The debanding algorithm doesn't work too well for blocking artifacts. Not as good as these test images seem to suggest. The debanding algorithm is really only meant to remove banding artifacts - not block artifacts.Yes, those are brightened a lot. Things like mosquito noise are compression artifacts, but blocking is really another form of banding. (which is why 10-bit encoding helps reduce it)

Can everybody live with this?After running through some more samples, I'm OK with this. I really would have liked to push maxAngle higher (0.18 really improves debanding in some cases) but the detail loss is too much with some sources.

[madshi]

Can I get an "ok" from bacondither, too? Then we'd have the LOW setting set.

So, medium and high settings now... :)

[turbojet]

I would say that's a candidate for low or medium settings, not high.

In that clip banding becomes a distraction to me below dif 1.5. I'm using pre and post sharpening which makes it stick out more, also been debanding for years so use to not seeing banding.

Is there any clip where grad 4.3 looks better than 2.3? Your deband samples didn't show any improvement to me but reduced details most noticeable in darker scenes. High gets knocked for lack of detail and grad is the value increased the most but is it too high?

[KeyserSoze17]

@madshi - I'm trying to get a direct3d device that I can use in a D3DImage and WPF.

I've tried creating new direct3D devices and then sending to madVR via the IMadVRDirect3D9Manager. It seemed like submitting the direct3d device the interface worked as I did not get an error code from that method. However, my application immediately crashed after that line of code.

Alternatively, I tried to get the madVR direct3d device using the IOsdRenderCallback and SetDevice API calls. I couldn't get that to work either.

My app (http://forum.doom9.org/showthread.php?t=167183) is coded in C# and I'm working with the madVR interface through ComImport functions, which in general is a challenge for me. I'm sure the madVR API is working properly and my issue is just trying to understand how the API is intended to work then how to translate that into a proper ComImport.

Do you have any tips on using these particular API interfaces? Do you have any sample code (does not need to be C#) showing the proper use of these particular interfaces?

Thank you for any assistance. And thank you for such an awesome piece of software.

[Werewolfy]

Might be time to look into "medium" and "high" settings. For medium I'm wondering whether we should increase all thresholds by a similar factor. Or whether we should maybe only increase thresholds slightly and use a wider "maxAngle" for the AngleBoost instead?

So any suggestions for "medium" and "high" anybody?


avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.0 2.8 1.8 1.2 off 1.9 0.18
high: 1.8 3.9 2.4 1.2 off 1.9 0.22


I didn't increase every values so much but with angleBoost and a wider maxAngle, do we really need higher values now? I think it does a really good job. I wouldn't increase gradient a lot more than this especially for medium because it can blur a lot of details.

[turbojet]

Is it just me or is angleboost about 5x heavier on the gpu than localcontrast?

[Ver Greeneyes]

madshi, I imagine this isn't exactly a long term thing, but could you make the text for the debanding filter settings wrap around or scale with the video window or something? On my 1366x768 laptop monitor, maxAngle gets cut off halfway even if I make the window as big as I can, so I can't see what it's set to.

[kasper93]

So my current favorite for LOW:

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 2.1 1.4 1.2 off 1.9 0.10
Can everybody live with this?

Great work! I've just tested this settings and indeed it's great balance between debanding and detail lost. You did awesome job here, thanks! And thanks to all who helped to pick those settings. I've tried to fine tune on my samples, but almost every time I end up with pretty the same results. maxAngle is low, but it's very good for low preset ;)

Or whether we should maybe only increase thresholds slightly and use a wider "maxAngle" for the AngleBoost instead?

After playing with different values I think this should be best approach, I will try tomorrow to find exact values. But generally I like pure image with very minor improvements, so I probably will not like > low ;p But for some sources this is needed.

[ryrynz]

Can I get an "ok" from bacondither, too? Then we'd have the LOW setting set.


Don't forget me, I'm still here.


So my current favorite for LOW:

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 2.1 1.4 1.2 off 1.9 0.10
Can everybody live with this?


Looks pretty good, a bit smooth though and that's all angleBoost, IMO it's too high.

The medium preset would be the place to crank things like angleBoost up, it affects quite a bit of the whole picture. Dropping it down brings back the detail without affecting debanding that much at all.

avgDif maxDif grad Penalty LC Boost maxAngle
low: 0.6 2.1 1.4 1.2 off 1.0 0.10



I have a few samples where a maxAngle of 0.10 seems a tad too low. I think 0.15 might be necessary?

I compared 0.01 directly to 0.25 and the differences were very subtle with the above settings, it did nothing to hurt or improve image quality so I'd almost be tempted to just drop it down to 0.01 here.

For the record I'm using the program Beyond Compare to show the relevant pixel changes between the preferences, I can compare the most affected changed areas rather easily.

[madshi]

I'm trying to get a direct3d device that I can use in a D3DImage and WPF.

I've tried creating new direct3D devices and then sending to madVR via the IMadVRDirect3D9Manager. It seemed like submitting the direct3d device the interface worked as I did not get an error code from that method. However, my application immediately crashed after that line of code.
The IMadVRDirect3D9Manager interface was added on request by tiben, when he tried to add support for madVR to XBMC. Since our efforts were "stopped", nobody is using this interface. It is possible that it isn't working as intended, although tiben did have some success with it, back in the day.

Do you get a proper crash report from madVR? Or does the crash occur outside of madVR?

Alternatively, I tried to get the madVR direct3d device using the IOsdRenderCallback and SetDevice API calls. I couldn't get that to work either.
PotPlayer is actively using this method, so it should work.

My app (http://forum.doom9.org/showthread.php?t=167183) is coded in C# and I'm working with the madVR interface through ComImport functions, which in general is a challenge for me. I'm sure the madVR API is working properly and my issue is just trying to understand how the API is intended to work then how to translate that into a proper ComImport.

Do you have any tips on using these particular API interfaces? Do you have any sample code (does not need to be C#) showing the proper use of these particular interfaces?
Unfortunately I've zero experience with C# or DotNet, so I fear I won't be of much help here... :( There was someone else trying to make madVR work with DotNet recently (you will find his posts if you browse through the last 20 pages or so), but I think he had problems already with more basic things, which you already seem to have solved, so his posts and my answers might not help you.

avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.0 2.8 1.8 1.2 off 1.9 0.18
high: 1.8 3.9 2.4 1.2 off 1.9 0.22


I didn't increase every values so much but with angleBoost and a wider maxAngle, do we really need higher values now? I think it does a really good job. I wouldn't increase gradient a lot more than this especially for medium because it can blur a lot of details.
Cool. Any comments anyone?

Is it just me or is angleboost about 5x heavier on the gpu than localcontrast?
It is not surprising that it might be a bit heavier because it requires an additional shader pass, which localContrast did not require. However, the amount of memory bandwidth should be more or less the same, so I'm not sure how much heavier angleBoost should really be. Anyway, it's the way it is, and I don't think there's any room for me to optimize this further.

madshi, I imagine this isn't exactly a long term thing, but could you make the text for the debanding filter settings wrap around or scale with the video window or something? On my 1366x768 laptop monitor, maxAngle gets cut off halfway even if I make the window as big as I can, so I can't see what it's set to.
I can't wrap or scale easily. I could maybe shorten some texts, but I'm not sure if I want to release a new test build just for that. You can see the values in the registry, if that helps?

HKCU\Software\madshi\madVR\maxAngle

It is updated at once in the registry when you press a key shortcut.

Great work! I've just tested this settings and indeed it's great balance between debanding and detail lost. You did awesome job here, thanks! And thanks to all who helped to pick those settings. I've tried to fine tune on my samples, but almost every time I end up with pretty the same results. maxAngle is low, but it's very good for low preset ;)
:)

After playing with different values I think this should be best approach, I will try tomorrow to find exact values.
Yes, please do!

Don't forget me, I'm still here.
I just mentioned the most active contributors, but your input is very welcome, too, of course!

Looks pretty good, a bit smooth though and that's all angleBoost, IMO it's too high.

The medium preset would be the place to crank things like angleBoost up, it affects quite a bit of the whole picture. Dropping it down brings back the detail without affecting debanding that much at all.

avgDif maxDif grad Penalty LC Boost maxAngle
low: 0.6 2.1 1.4 1.2 off 1.0 0.10


I compared 0.01 directly to 0.25 and the differences were very subtle with the above settings, it did nothing to hurt or improve image quality so I'd almost be tempted to just drop it down to 0.01 here.
Basically you've already turned the angleBoost feature completely off! "angleBoost = 1.0" means that it doesn't boost anything at all, it's identical to turning it off.

I'm not sure if I like your settings. The problem is that angleBoost aims at improving the "debanding strength vs detail loss" equation. So if you disable this feature, of course that means less detail loss, but it also means even more loss of debanding strength. Instead if might be more effective to lower the other thresholds even further and activate angleBoost instead. That way the detail loss would be even smaller than with your suggested settings in most situations, while debanding strength would be equal or even slightly better. E.g. Werewolfy suggested something like that here:

http://forum.doom9.org/showpost.php?p=1648737&postcount=20456

[ryrynz]

Basically you've already turned the angleBoost feature completely off! "angleBoost = 1.0" means that it doesn't boost anything at all, it's identical to turning it off.


I tested 1.0 vs off and I've just done so again now and there's a difference between them, are you sure it's actually being disabled at 1.0?

I find when I take screenshots there is always a slight variation in the images even with using the same settings and I don't know why, but even so I can't see any differences. I can however see a difference between angleboost 1.0 and off.

[tobindac]

Often the the player getting stuck for several seconds or hanging when used in conjunction with SVP, especially when skipping back. Any idea how to remedy it? Radeon 6570M on a Windows 8.1 (or 8) 64.

[madshi]

I tested 1.0 vs off and I've just done so again now and there's a difference between them, are you sure it's actually being disabled at 1.0?

I find when I take screenshots there is always a slight variation in the images even with using the same settings and I don't know why, but even so I can't see any differences. I can however see a difference between angleboost 1.0 and off.
Every time you change any rendering settings, the dithering pattern changes, because it's random. I think that this is probably what you're seeing. The AngleBoost feature works by multiplying the other thresholds (avgDif, maxDif and gradient) with the "angleBoost" value. So an AngleBoost value of 2.0 doubles the avgDif, maxDif and gradient thresholds, if the gradient angle is detected as being similar. Of course avgDif * 1.0 doesn't change anything. That's why AngleBoost = 1.0 shouldn't show any changes. "angleBoost" doesn't change the pixels directly. All it does is change the other thresholds. Unless I've a bug in my algorithm, but I don't think so.

Often the the player getting stuck for several seconds or hanging when used in conjunction with SVP, especially when skipping back. Any idea how to remedy it? Radeon 6570M on a Windows 8.1 (or 8) 64.
No idea, I'm sorry. Are you sure it's not an issue with SVP?

[ryrynz]

Unless I've a bug in my algorithm, but I don't think so.

It looks like I'm just noticing the dithering pattern changes.

Instead if might be more effective to lower the other thresholds even further and activate angleBoost instead.

I'll look into it, thanks.

[iSunrise]

madshi, I just wanted to do some testing with your colors.ytp and grayramp.ytp files, where I would´ve expected to see smooth gradients, but instead I see big blocks. Was is always like that? I kinda remember it being smooth?! Am I imagining things?

This is with your latest test build btw.

[madshi]

It was always that way, and intentionally. Those "blocks" you see in the grayramp.ytp test pattern are actually 8bit 1 value changes. That just goes to show that 8bit is not enough to produce a smooth ramp. Thankfully the new debanding algorithm nicely takes care of that...

[iSunrise]

It was always that way, and intentionally. Those "blocks" you see in the grayramp.ytp test pattern are actually 8bit 1 value changes. That just goes to show that 8bit is not enough to produce a smooth ramp. Thankfully the new debanding algorithm nicely takes care of that...
Ok, thanks. I thought something on my side is messed up.

Yes, the debanding does an incredible job of smoothing them out. The blockiness is completely gone.

[ryrynz]

Can everybody live with this?



I'd like to see maxDif decreased a touch, thanks.

avgDif maxDif grad Penalty LC Boost maxAngle
average: 0.6 1.9 1.4 1.2 off 1.9 0.10

[Werewolfy]

Have you found find a video where maxDif blurs some details? To be honest, I've never seen a real negative side effect for this one so it interests me if you've found one.

[Ver Greeneyes]

You can see the values in the registry, if that helps?

HKCU\Software\madshi\madVR\maxAngle
That's fine, thanks!

[leeperry]

Anyone use the BT.709 Gamma profiles, and think they look like they got elevated blacks and too low gamma (1.95)?

That's because 709 gamma specs are for Encoding ONLY !!!
NOT for calibrating displays, period.
Well, I sure hope it won't be removed because I really enjoy it when set to 2.2 as it nicely compensates for the dark gamma preset I am forced to set in my TV's OSD.

Too bad the darn thing isn't natively running D65 but more like 6800K so I'll be forced to build a D65/BT.1886 LUT but still, it's a one-button tweak that looks great to me and not every mVR user will care to use a meter.

[Fadeout]

MadVR + shaders + adjustments to video size through Pan&Scan produce crashes at certain points of videos (and only in full screen exclusive mode).

Can anyone figure out why and how to fix?

[shimaflarex]

MadVR + shaders + adjustments to video size through Pan&Scan produce crashes at certain points of videos (and only in full screen exclusive mode).

Can anyone figure out why and how to fix?

Are you using XySubFilter?
It may be related to this bug: http://code.google.com/p/xy-vsfilter/issues/detail?id=153

[Fadeout]

Are you using XySubFilter?
It may be related to this bug: http://code.google.com/p/xy-vsfilter/issues/detail?id=153

Yeah, that's it. Or at least I tried with subtitles off and it doesn't crash.

Are there nightly builds or something?

[shimaflarex]

Yeah, that's it. Or at least I tried with subtitles off and it doesn't crash.

Are there nightly builds or something?

Here: https://github.com/mpc-hc/mpc-hc/pull/114
If you are using MPC-HC, you may have to not only download XySubFilter but also the special build MPC-HC also located on that page (I did not test with a regular build).

[Soukyuu]

Seeing all the progress around debanding is great, I just hope that it won't encourage certain groups to encode material at much lower bitrates expecting madvr to compensate for it.

[ryrynz]

Have you found find a video where maxDif blurs some details? To be honest, I've never seen a real negative side effect for this one so it interests me if you've found one.

It's the fuzz on the peach as it were, looking to get back a bit touch of sharpness without sacrificing deband performance, it's all quite subtle.

[noee]

Anybody who needs the COM import and "includes" for madVR in DotNet, I've got a barebones+ little sample app with defs. PM me.

[ryrynz]

Seeing all the progress around debanding is great, I just hope that it won't encourage certain groups to encode material at much lower bitrates expecting madvr to compensate for it.

I wouldn't expect that to happen. Nobody should be expecting anyone to run specific configurations. H.265 uptake should be pretty quick helping in the bitrate area.
Does anyone feel it worthwhile having madvr avoid debanding for 10 bit content? Or at least an option to specify a lower preset for it?

[dansrfe]

If I understood the spec correctly, 10 bit shouldn't need debanding since that what it aims to solve with respect to 8 bit in the first place.

[shimaflarex]

Encoding on 10bit doesn't magically get rid of banding of the original source (which is usually 8 bit).

[dansrfe]

Encoding on 10bit doesn't magically get rid of banding of the original source (which is usually 8 bit).

Of course. I was talking about 10 bit sources around the time when HEVC becomes mainstream.

[smoke070]

Question for people with ATI cards.. when installing my drivers should I enable or disable these options? I'm wondering if these things interfere with madVR..

http://i.imgur.com/70pBNTT.jpg

[agustin9]

I have them disabled, those are not needed

[Asmodian]

They also do not conflict with or effect madVR in anyway. That is fairly low quality video transcoding though; I wouldn't use it so no reason to install it.

[ryrynz]

Have you found find a video where maxDif blurs some details? To be honest, I've never seen a real negative side effect for this one so it interests me if you've found one.

Compare these, you'll see maxdif at 2.1 is slightly softer. It's not a negative side effect, just a preference for a less soft appearance.

2.1
http://imageshack.us/a/img189/4088/878x.th.png (http://imageshack.us/photo/my-images/189/878x.png/)

1.9
http://imageshack.us/a/img571/5586/e5rp.th.png (http://imageshack.us/photo/my-images/571/e5rp.png/)

[vivan]

ryrynz,
those images a bit-identical...

[ryrynz]

Dunno what happened there, weird, will fix it up.