madVR - high quality video renderer (GPU assisted)

[ashlar42]

Originally Posted by ionutm80

So it's enough to switch tp D3D9 w/o disabling FSE?

In my experience, that's the case, yes.

[Werewolfy]

Originally Posted by XTrojan

Is it worth using SuperRes Anti-ringing and Bloat? With the new deringing and anti-halo, they take up the resources I use for SuperRes ringing, which one is worth more? I overall disliked de-haloing effects though.

I find anti-Bloating with SuperRes useless, it doesn't do any good to the picture. Anti-ringing with SuperRes has some small benefits so if you have the ressources, use it.

[Q-the-STORM]

Originally Posted by ionutm80

You lost me here, why am I losing frames? The original 24 should still be there, is just that I see more frames now (the computed and interpolated ones), or you are referring to the fact that I still do not have a perfect multiple of the original frames by an integer number?

Sorry to tell you, but one of the huge drawbacks of using SVP at 60Hz is that you actually lose every second original frame...

on order to interpolate to 60fps without jerky playback, you have to interpolate to 120fps and drop every second frame... so this is what happens:

(o=original i=interpolated, numbers=framenumbers)

o01 o02 o03 o04 o05 o06

you got these original 6 frames....
what SVP does is adding 4 frames in between every frame

o01 i01 i02 i03 i04 o02 i05 i06 i07 i08 o03 i09 i10 i11 i12 o04 i13 i14 i15 i16 o05 i17 i18 i19 i20 o06 i21 i22 i23 i24

so 6 frames become 30 frames, which means 24 frames become 120 frames... but since you play back at 60Hz, you have to drop every second frame... which means this is what you get:

o01 i02 i04 i05 i07 o03 i10 i12 i13 i15 o05 i18 i20 i21 i23

with every second frame dropped, 30 frames turn to 15 frames, so 120 frames turn to 60 frames...
but now half of the original frames are missing (o2 o4 and o6)...

if they only dropped interpolated frames when needed, you would get jerky playback...

that's why it's called 5:2 interpolation, you mulitply by 5, then divide by 2...
that's the reason the next TV I'm buying has to accept 120Hz signal, that way SVP can just multiply by 5 and not have to drop frames...

[huhn]

think about 48 fps and smoothmotion.

[ionutm80]

Originally Posted by Q-the-STORM

Sorry to tell you, but one of the huge drawbacks of using SVP at 60Hz is that you actually lose every second original frame...

on order to interpolate to 60fps without jerky playback, you have to interpolate to 120fps and drop every second frame...

Wow, thanks for clarifying ... I did not know that. So SVP is worth only if the TV can accept 120Hz?

[ionutm80]

Originally Posted by huhn

think about 48 fps and smoothmotion.

You are saying that I should use SVP at 48 fps and also activate smoothmotion in madVR? I was told that they are redundant and with SVP, smoothmotion is not needed ...

[huhn]

Originally Posted by ionutm80

Wow, thanks for clarifying ... I did not know that. So SVP is worth only if the TV can accept 120Hz?

you should ask that your eyes.

Originally Posted by ionutm80

You are saying that I should use SVP at 48 fps and also activate smoothmotion in madVR? I was told that they are redundant and with SVP, smoothmotion is not needed ...

it all depends on what you are doing. this setup isn't removing original frames.

[XTrojan]

Interpolating with SVP instead of using built in TV interpolation is just a bad idea, SVP has far more artifacts. SVP is more for monitors that can reach 120hz not TV screens.

There's also TVs that can reverse telecine or something similar if you want to play 24p movies at 60p with no issues, afaik all modern sony TVs can do this so go buy one of those.

[rivera]

Could anyone give me an advice?

Currently I am using GTX580 with madVR & 1080p TV.
For 1080p remuxes, 720p rips & 4K samples, I see no drops with these settings (ver. 0.90.19):
--------------------------------------------------------------------------------------
chroma upscaling:
NNedi3-32
activate SuperRes filter, strength=4

image downscaling:
SoftCubic, softness=70
activate anti-ringing filter

image doubling=OFF

image upscaling:
Jinc
activate anti-ringing filter

upscaling refinement:
crispen edges=1.1
enhance detail=1.2
AdaptiveSharpen=0.7
activate anti-ringing filter
--------------------------------------------------------------------------------------

Now I need to set up another PC with madVR (Z170 mobo, i3-6100).
This PC will be used with Panasonic plasma (PR65VT60) with 3DLut.
And I was going to use GTX960 graphics card (rather cold and quite for HTPC).

My questions are:
1. Will the new card (GTX960) be able to work with at least the same madVR settings without drops? (but without 3DLut)
2. Same question as above but with 3DLut.

Thanks in advance for answers.

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

Actually, I suspect I do not understand madVR settings properly.
For example, may be I do not need NNedi at all.
See, I have 1080p plasma TV, my content is 95% 1080p remuxes, and other 5% are 720p & 4K.
May be for that I need some other settings?

[Q-the-STORM]

well chroma always needs to be upscaled, so NNEDI3 is definitely beneficial... though on 720p content, you should use nnedi3 on image upscaling instead of chroma... that makes a much bigger difference... in chroma you should only use nnedi3 if you have maxed out all other settings, since it barely makes a difference to super-xbr...

btw making profiles for each resolution (SD, 720p, 1080p, 2160p) is always good, doesn't take that long to test settings for each one with no frame drops...
Just create a profile group in scaling algorithms and add 4 profiles with all scaling algorithm entries (chroma upscaling, image upscaling etc)

This is what my auto select rules look like on my profile group

if (srcWidth < 950) and (srcHeight < 700) "SD"
if (srcWidth >= 950) and (srcWidth <= 1420) and (srcHeight >= 700) and (srcHeight <= 740) "720p"
if (srcWidth < 2000) and (srcHeight < 1100) "1080p"
else "2160p"

that way profiles will be selected based on resolution and you can tweak settings for each resolution...
that way you can have different settings for each resolution, makes a lot more sense....

[madshi]

Originally Posted by ajp_anton

Small request:

Current option:
"disable scaling if image size changes by only:"
Could you add a sub-option below it:
"allow integer-factor resizes to hit the same target"

to have for example a 1918x1078 video upscale to 3836x2156, for performance reasons.

Such an option would be very hard to understand for users. An option like that belongs would also better fit into the "trade quality" section. Anyway, atm I don't want to do this. Maybe in some future version.

Originally Posted by CarlosCaco

if i set supersamplig image doubling activate but upscalling refinement doesn t activate

Ok. So how can I reproduce this problem? I'll need a step-by-step guide. Meaning: I need your "settings.bin" file (or "HKCU\Software\madshi\madVR\Settings", if there's no "settings.bin" on your PC). Then I need to know the exact source and target resolutions we're talking about. And a debug log. All this from the situation where image doubling activates but upscaling refinement does not.

Originally Posted by _rEuTeL_

But could you please fix the bug with italic subs that breaks the 'move subs to bottom of the screen' option? It's the only issue I have with MadVR after months of use

As I've said many times before, it's on my to do list. But it needs to be fixed in XySubFilter, which is very painful for me to do. So I don't know when I'll get to that.

Originally Posted by XMonarchY

I was HOPING madVR could MOVE subtitles to the very bottom black bar (for 21:9 content format, which this foreign film used), but no matter which settings I adjusted, I couldn't move the subtitles anywhere. The subtitles were integrated into the ripped MKV file itself and were not coming from a separate subtitles file. I know the subtitles were not hard-coded because I could turn them off completely...

- Can madVR move non-hard-coded subtitles to the bottom of the screen, where black bar is located in 21:9 film content?
- If madVR CAN move the subtitles to the bottom of the screen (on the black bar), then which settings do I need to adjust and which values should I use for those settings?

That depends on the subtitle format (e.g. text vs bitmap) and the subtitle renderer (internal MPC-HC/BE vs XySubFilter) you're using.

Originally Posted by omarank

Upon toggling the two states of "linear" LUTs (65280 ending and 65535 ending) while playing some movie content, I find the images look better (more natural) when the LUT is in the OS default state (ending at 65280). But this may be subjective.

I wish madVR could completely disable the LUT processing instead of setting the LUT linear. In that case even when the LUT is loaded, madVR could read the LUT and apply it internally in high bit depth like it does for Overlay mode.

I may do that in some future version, but it's low priority for now.

Originally Posted by rivera

1. Will the new card (GTX960) be able to work with at least the same madVR settings without drops? (but without 3DLut)
2. Same question as above but with 3DLut.

I've no idea. Check some gaming benchmarks. If the 960 is faster in games than the 580 then it should probably work.

[madshi]

madVR v0.90.20 released

http://madshi.net/madVR.zip

* tweaked deringing algorithm some more
* added workaround for weird crash with latest AMD beta drivers

Not much time for madVR this weekend, so only small changes. Feedback about the tweaked deringing algo welcome.

[aufkrawall]

Thanks for the changes, is there now a grain detection implemented for deringing?

Result in this cartoon looks very clear and also otherwise well to me. Look at how enormous the difference with super-xbr sharpness 150 is:
off:


on:


Looks like pure magic is going on.

[nghiabeo20]

A question from a newbie here: I've just have my laptop calibrated, but from what I've read, madvr bypass all gpu processing, just use it's own algorithm. So will madvr disable icm profile? Thanks!

[huhn]

no it doesn't bypass that by default.

windows loaded icm calibration is not the best way to calibrate.
think about creating 3d LUT from the icm using this: http://displaycal.net/

or use overlay mode if possible.

[JarrettH]

Is that worth using if I already calibrate with my Spyder2 software? I've heard of it, but it seems even more technical. Is it better?

[huhn]

yes the spyder and iprofiler software is not that good to say it friendly.

[leeperry]

OMG yet another new build,

BTW, I recently got ahold of a DVB-T2 tuner that can record on USB keys so I finally get to use the black bars detection but:

1) could you please add tags for borders cropping? some TV logos only need a few lines of cropping to fully disappear and I'd like to nail this once and for all.

3) hardly worth a heart attack but why does the latter jump from 5 to 7? Eventually I got a video that would ideally need 6 ^^

4) TV broadcasters like to mess with levels and even more AR, there's no tag for the latter according to this as it would appear? I've seen many 2.35 music videos airing as 1.78

Due to the new deringer I'm contemplating watching everyday TV through mVR but I can't quite justify wasting +100W to watch the news and the usual TV junk, maybe things will change with the next AMD boards and make it far more worth my while but YADIF in LAV gives combing artifacts on 1080i50@25p fast moving logos IME and I can't process 50p. My Sammy TV hardly ever combs, would need to try mVR's deinterlacer I guess.

Also, switching fps in Reclock or closing PotP takes a few secs now when it used to be instant before so I will try older builds to see when that really started and whether mVR is even the culprit.

[leeperry]

Sorry for the double post and I do realize that I'm extremely late to the party but I think a few more lines of cropping woulda been in good order?



Please advise whether that'd make a sample useful to anything as many music videos on that channel end up with that white bar in the top right corner

[Asmodian]

Originally Posted by JarrettH

Is that worth using if I already calibrate with my Spyder2 software? I've heard of it, but it seems even more technical. Is it better?

Especially the Spyder2 is too old for modern displays, I had to replace my Spyder3 years ago because it could not read accurate colors on the new WLED backlights that seem ubiquitous now or on anything with a wide gamut. Also as they age the filters change so the meters do simply go bad with time (heat, water, oxygen, etc.). Edit: I used the meter with Argyllcms, not their software, which as mentioned, is not very good.

A 3DLUT and a Windows calibration as done by Spyder's software is not the same thing. Think of the Windows calibration as the first step of a calibration and the 3DLUT as the full calibration done in one step.

huhn mentioned using the rendering option "enable windowed overlay" because madVR bypasses the GPU's calibration hardware completely in that mode but it loads the values from the GPU and applies then internally at its high bit depth and before dithering. This does offer better quality but it does not help if the values are bad to start with.

Originally Posted by leeperry

Please advise whether that'd make a sample useful to anything as many music videos on that channel end up with that white bar in the top right center

Is the "cleanup image borders by cropping" option not a good solution? It is probably meant to clean up faded or distorted edges of analogue captures but it could work well here too?