Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Hardware & Software > Software players

Reply
 
Thread Tools Search this Thread Display Modes
Old 8th May 2016, 13:38   #37801  |  Link
Xaurus
Registered User
 
Join Date: Jun 2011
Posts: 288
I've tried to read through the last number of pages and looked at some guides, but I can't figure out how to let madvr treat movie clips that have "odd" resolution, like 1916 x 1076.
I am upscaling everything to 3840 x 2160. A normal 1920 x 1080 movie has roughly 6-7 ms "to spare" relative to the frame interval of 41.67 ms, but these movies with 1916 x 1076 almost eats up this buffer and brings the rendering time closer to 41.67 ms.

One way is of course to add a profile for each such "odd" resolution, but I looked in zoom control and checked the "disable scaling if image size changes by only" option to "5 lines or less".
MPC-HC is set to "touch window from inside". But there is absolutely no difference what so ever. Am I misunderstanding this option?

Would appreciate any input relative to handling these "odd" format movies.
__________________
SETUP: Win 10/MPC-HC/LAV/MadVR
HARDWARE: Fractal Design Node 804 | Xeon E3-1260L v5 | Supermicro X11SSZ-TLN4F | Samsung 2x8GB DDR4 ECC | Samsung 850 EVO 1TB | MSI GTX 1650 Super | EVGA G2 750
Xaurus is offline   Reply With Quote
Old 8th May 2016, 14:06   #37802  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,890
you are scaling for 1924 lines. so you can't use that option.
huhn is offline   Reply With Quote
Old 8th May 2016, 16:32   #37803  |  Link
alf6
Registered User
 
Join Date: Feb 2016
Posts: 4
Quote:
Originally Posted by huhn View Post
reproduced.

lanczos is used or at least shown in the OSD when the image is only scaled in X or Y direction.

jinc is correctly used with a 1440x1080p 16/9 AR source displayed at 1920x1080p.
there is an additional "image y = jinc" line.

you have set your player to stretch. set it to "touch window from the inside" (name can depend on the player) if you don't want that.
Colud anyone recommend how to fix this in mpc-hc?
alf6 is offline   Reply With Quote
Old 8th May 2016, 17:10   #37804  |  Link
Xaurus
Registered User
 
Join Date: Jun 2011
Posts: 288
Quote:
Originally Posted by huhn View Post
you are scaling for 1924 lines. so you can't use that option.
Hmm can't say I am following you.

Just answer me this: is there anything in madvr settings that can deal with "odd" resolutions like 1916 x 1076 except for creating a new profile for it?
__________________
SETUP: Win 10/MPC-HC/LAV/MadVR
HARDWARE: Fractal Design Node 804 | Xeon E3-1260L v5 | Supermicro X11SSZ-TLN4F | Samsung 2x8GB DDR4 ECC | Samsung 850 EVO 1TB | MSI GTX 1650 Super | EVGA G2 750
Xaurus is offline   Reply With Quote
Old 8th May 2016, 19:14   #37805  |  Link
baii
Registered User
 
Join Date: Dec 2011
Posts: 180
Quote:
Originally Posted by Xaurus View Post
Hmm can't say I am following you.

Just answer me this: is there anything in madvr settings that can deal with "odd" resolutions like 1916 x 1076 except for creating a new profile for it?
Do you want to have black bar for those odd resolutions to trade for better performance?

Sent from my 306SH using Tapatalk
baii is offline   Reply With Quote
Old 8th May 2016, 20:01   #37806  |  Link
Xaurus
Registered User
 
Join Date: Jun 2011
Posts: 288
Quote:
Originally Posted by baii View Post
Do you want to have black bar for those odd resolutions to trade for better performance?

Sent from my 306SH using Tapatalk
If that means a tiny black bar for width and for height, sure thing. I don't know how though, I've played around with zoom control but the performance stays the same no matter what I do...
__________________
SETUP: Win 10/MPC-HC/LAV/MadVR
HARDWARE: Fractal Design Node 804 | Xeon E3-1260L v5 | Supermicro X11SSZ-TLN4F | Samsung 2x8GB DDR4 ECC | Samsung 850 EVO 1TB | MSI GTX 1650 Super | EVGA G2 750
Xaurus is offline   Reply With Quote
Old 8th May 2016, 22:57   #37807  |  Link
dansrfe
Registered User
 
Join Date: Jan 2009
Posts: 1,210
Quote:
Originally Posted by Xaurus View Post
If that means a tiny black bar for width and for height, sure thing. I don't know how though, I've played around with zoom control but the performance stays the same no matter what I do...
Make sure you have crop enabled in order to remove the black bars from the processing chain and get a potential render time improvement.
dansrfe is offline   Reply With Quote
Old 8th May 2016, 23:24   #37808  |  Link
Xaurus
Registered User
 
Join Date: Jun 2011
Posts: 288
Quote:
Originally Posted by dansrfe View Post
Make sure you have crop enabled in order to remove the black bars from the processing chain and get a potential render time improvement.
What exactly must I set in order to get this working?

So I have set:
"disable scaling if image size changes by only ... 5 lines or less"
"move subtitles to bottom of the screen/window"
"automatically detect hard coded black bars"
"crop black bars"

This doesn't make any difference...
Should MPC-HC stay at "touch window from inside" or some other setting?

Scaling info with a normal 1920x1080 @ 3840x2160:
chroma > Jinc AR
luma > Nnedi 64
chroma > Jinc AR
Rendering times ~35 ms

Scaling info with 1916x1076 @ 3840x2160:
chroma > Jinc AR
luma > Nnedi 64 > Jinc AR
chroma > Jinc AR
Rendering times ~40 ms

As you can see it adds roughly 5 ms to the rendering time which I what I want to avoid.

Thanks.
__________________
SETUP: Win 10/MPC-HC/LAV/MadVR
HARDWARE: Fractal Design Node 804 | Xeon E3-1260L v5 | Supermicro X11SSZ-TLN4F | Samsung 2x8GB DDR4 ECC | Samsung 850 EVO 1TB | MSI GTX 1650 Super | EVGA G2 750
Xaurus is offline   Reply With Quote
Old 8th May 2016, 23:38   #37809  |  Link
bill4
Registered User
 
Join Date: Feb 2016
Posts: 3
Can we expect DX12 and/or Vulcan support at some point in the future?
bill4 is offline   Reply With Quote
Old 9th May 2016, 01:09   #37810  |  Link
JarrettH
Registered User
 
Join Date: Aug 2004
Location: Canada
Posts: 860
Quote:
Originally Posted by Xaurus View Post
What exactly must I set in order to get this working?

So I have set:
"disable scaling if image size changes by only ... 5 lines or less"
"move subtitles to bottom of the screen/window"
"automatically detect hard coded black bars"
"crop black bars"

This doesn't make any difference...
Should MPC-HC stay at "touch window from inside" or some other setting?

Scaling info with a normal 1920x1080 @ 3840x2160:
chroma > Jinc AR
luma > Nnedi 64
chroma > Jinc AR
Rendering times ~35 ms

Scaling info with 1916x1076 @ 3840x2160:
chroma > Jinc AR
luma > Nnedi 64 > Jinc AR
chroma > Jinc AR
Rendering times ~40 ms

As you can see it adds roughly 5 ms to the rendering time which I what I want to avoid.

Thanks.
Check "disable scaling if image size changes by" (pick 10 lines or less)

Edit: And check "automatically detect hard coded black bars" with "crop black bars"

...sorry I see you have both options on, but I don't think 5 lines or less is sufficient

In the latter statistics that is happening because image doubling doesn't QUITE fill a 4k screen; in fact, if you choose 10 lines or less that should fix it!

Last edited by JarrettH; 9th May 2016 at 01:24.
JarrettH is offline   Reply With Quote
Old 9th May 2016, 01:38   #37811  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,890
Quote:
Originally Posted by JarrettH View Post
Check "disable scaling if image size changes by" (pick 10 lines or less)

Edit: And check "automatically detect hard coded black bars" with "crop black bars"

...sorry I see you have both options on, but I don't think 5 lines or less is sufficient

In the latter statistics that is happening because image doubling doesn't QUITE fill a 4k screen; in fact, if you choose 10 lines or less that should fix it!
nope the scaling factor is still 1924 lines not 4 lines.

even if you set it to 50 lines it will not stop scaling. i tried it already.
huhn is offline   Reply With Quote
Old 9th May 2016, 01:58   #37812  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by Xaurus View Post
What exactly must I set in order to get this working?

So I have set:
"disable scaling if image size changes by only ... 5 lines or less"
"move subtitles to bottom of the screen/window"
"automatically detect hard coded black bars"
"crop black bars"

This doesn't make any difference...
Should MPC-HC stay at "touch window from inside" or some other setting?

Scaling info with a normal 1920x1080 @ 3840x2160:
chroma > Jinc AR
luma > Nnedi 64
chroma > Jinc AR
Rendering times ~35 ms

Scaling info with 1916x1076 @ 3840x2160:
chroma > Jinc AR
luma > Nnedi 64 > Jinc AR
chroma > Jinc AR
Rendering times ~40 ms

As you can see it adds roughly 5 ms to the rendering time which I what I want to avoid.

Thanks.
Try setting image upscaling to Lanczos3. You're just over 2x scaling.

Last edited by Warner306; 9th May 2016 at 03:00.
Warner306 is offline   Reply With Quote
Old 9th May 2016, 08:49   #37813  |  Link
MokrySedeS
I am the one who knocks
 
MokrySedeS's Avatar
 
Join Date: Aug 2009
Posts: 104
I thought that maybe executing a command line in a profile with this script (set mpc-hc to 200% instead of touch window from inside) would solve it for Xaurus but sadly this option is broken in madVR http://bugs.madshi.net/view.php?id=210
MokrySedeS is offline   Reply With Quote
Old 9th May 2016, 08:54   #37814  |  Link
nevcairiel
Registered Developer
 
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,336
Quote:
Originally Posted by bill4 View Post
Can we expect DX12 and/or Vulcan support at some point in the future?
Both of these APIs are not particularly useful for video rendering, as you only render one thing, a video image, not a million different objects on screen.
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
nevcairiel is offline   Reply With Quote
Old 9th May 2016, 09:46   #37815  |  Link
regiregi22
Registered User
 
Join Date: May 2014
Posts: 22
I have a high-end CPU and a mid-high-end Nvidia GPU. I am wondering what would be the most efficient approach regarding load balancing.

If I am not wrong, I recall that madvr uses CUDA from the GPU for rendering. Would it be wise to use "None" has hardware decorder, so the CPU is used for decoding while the whole GPU power is spared to madvr?

Or, on the other hand, if madvr uses the CPU, would it be better to use CUDA hardware decoding and leaving the whole CPU for madvr?
regiregi22 is offline   Reply With Quote
Old 9th May 2016, 10:08   #37816  |  Link
nevcairiel
Registered Developer
 
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,336
madVR doesn't use the CPU much, so if its fast enough to decode everything you encounter, using software decoding is generally a good option.
Note that CUDA decoding is not generally recommended anymore, and if you go with hardware decoding, favor DXVA2 (copy-back) instead.
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
nevcairiel is offline   Reply With Quote
Old 9th May 2016, 10:44   #37817  |  Link
bill4
Registered User
 
Join Date: Feb 2016
Posts: 3
Quote:
Originally Posted by nevcairiel View Post
Both of these APIs are not particularly useful for video rendering, as you only render one thing, a video image, not a million different objects on screen.
Thanks. I just thought Vulcan (as it's closer to the hardware) could have been faster. Thanks anyway.

As far as DX12 is concerned, the advantage would be to use 2 different cards (AMD & nVidia) together to increase available power.
bill4 is offline   Reply With Quote
Old 9th May 2016, 11:33   #37818  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,890
Quote:
Originally Posted by bill4 View Post
Thanks. I just thought Vulcan (as it's closer to the hardware) could have been faster. Thanks anyway.

As far as DX12 is concerned, the advantage would be to use 2 different cards (AMD & nVidia) together to increase available power.
it's not that simple.

madVR is working with huge images and both cards need these images.

PCI-e is simply not fast enough to get these images from one GPU to any other GPU.
video processing feature usually need the whole image to get a proper results so splitting the image doesn't work too.
huhn is offline   Reply With Quote
Old 9th May 2016, 13:28   #37819  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by mogli View Post
On what is it based then?
Can they be compared like 'super-xbr anti-bloating 100' meaning 'super-xbr 150 minus the softening anti-bloating 100 does' or something?
Simply consider the anti-bloating versions of super-xbr as slightly different algorithms, with different parameters.

Quote:
Originally Posted by chros View Post
if I check 'scale chrome separately, if ...' then indeed chroma conversion vanishes (so maybe this new option brought this bug, if it's a bug)
Quote:
Originally Posted by nevcairiel View Post
Well sounds to me like this option is designed to allow just what you want, scale chroma straight to the target (which in this case is no scaling at all).
^ It's as nevcairiel said. It's not a bug. The behaviour is exactly what the trade quality option is meant to do.

Quote:
Originally Posted by har3inger View Post
Superres with antibloat at or above 100% completely eliminates the aliasing/blobbiness/chunkiness SR introduces to low resolution content. Unfortunately, it also makes it appear that superres is doing almost nothing except introduce artifacts on hardcoded subs and similar scene details.

For low-res bad quality content, SXBR AB25 looks quite similiar to nnedi64. There's a bit less sharpness, but similar reduction in source artifacts. Considering the cost, that's pretty damn good. Any higher AB and SXBR starts looking too soft. Using AB on SXBR followed by SR with AB checked can tend to produce odd looking images.
K, thanks.

Quote:
Originally Posted by har3inger View Post
What quality tradeoffs are present when we check "scale chroma separately"?
Unchecking that option always upscaled chroma first, then converts to RGB, then scales the RGB image to the final resolution. Checking that option scales Y and the CbCr channels separately, if doing so improves performance. Quality is slightly better when always upscaling chroma first.

Quote:
Originally Posted by Warner306 View Post
Would it be possible to start a thread where test images are available for download? By that, I mean the unaltered, original image. I am new to this but have discovered the following:
  • Not every image is a good test image;
  • Good test images are difficult to find on the Web;
  • Most images only show differences in parts of the image, not the entire image;
  • Some images show no difference at all;
  • Images must be the correction resolution and size to allow for adequate resizing.
If test images are indeed the best way to assist this enterprise, good test images should be available to everyone interested in testing.
I don't have the time for that, but if you want to do that, that's fine with me. I have a couple of test images I like to use, you can find them in my posts in this thread.

Quote:
Originally Posted by mzso View Post
I don't see how "Ctrl+Alt+1" is relevant. That's neither "alt+[something]" nor does it include a character with a diacritic mark (like these: íöüóőúáűé)
I tried Ctrl+Alt+Q/W, which works here also. But not with only Alt as a modifier.
At some point I'm going to improve the keyboard shortcut stuff. Instead of having to type in the "name" of a key, you will be able to simply press the key you want and madVR will "record" it. I suppose that should solve this problem you're reporting. Sorry, will have to wait until I get to that, which could be a while...

Quote:
Originally Posted by huhn View Post
i tried anti bloat 25-75 now and i can't find something that is useful for pure up scaling.

it kind of creates on effect of double lining where anti bloat 100 was so unsharp i didn't really notice this.

FHD->UHD with crisp edges 1.5 upscale refinment is already clearly highlighting this effect. and that for all anti bloat super xbr settings.

even super xbr anti bloat 25 isn't as sharp as super xbr 100 and every higher anti bloat setting isn't as sharp as spline 3.

so here is the main issue upscale refinement is needed with super xbr anti bloat which has issues with crisp edges.
Ok, I'll see if I can modify the super-xbr anti-bloating settings to make all those options a bit sharper. Of course that will also decrease the anti-bloating effect a bit.

Quote:
Originally Posted by krille View Post
Don't know if this has been reported, but since version 0.90.16 there is a problem with dxva deinterlacing for me. The picture is extremely pale when deinterlacing is active.
I'm using Nvidia 960 with latest driver.
I'll need more details. E.g. which decoder are you using? Software or hardware? Does anything change if you modify the decoder settings? Have you tried with madVR default settings? What happens if you toggle the "use DXVA chroma upscaling" options in the madVR "trade quality" settings section?

Quote:
Originally Posted by markanini View Post
For hires photos SSIM downscaling w/ AR+AB100% seemed like the best compromise. I couldn't see any obvious differences between relaxed and strict AR. I vote for whatever uses the least GPU cycles.
Quote:
Originally Posted by markanini View Post
I think I can now retire NEDI+SR and replace it with "super-xbr, anti-bloating:75". It's almost as good as NEDI+SR at minimizing blocking and gibbs artefacts and will use much less GPU. Note: I always have enhance detail: 1.0 pre-resize.
Ok, thanks for the feedback.

Quote:
Originally Posted by leeperry View Post
Antibloat appears too strong in the new sxbr doubling options.
I don't really understand what you mean. Instead of trying to describe which technical aspect of the algorithm could be problematic, try describing what you don't like about the final image quality when using super-xbr with anti-bloating.

Quote:
Originally Posted by Werewolfy View Post
Yes, it's useful. I find AdaptiveSharpen to look more natural with anti-bloating 75% or 100%.

I only tested the anti-bloating filter so far and I don't like it for SuperRes, it removes all the benefits brought by SuperRes even with low values.
Thanks.

Quote:
Originally Posted by Neo-XP View Post
"ldmp" because it's set too high in madVR and it forbids the enhancement of very small details. When set too low, the noise is sharpened too, but a noise remover can be used to avoid this.
ldmp is set high intentionally because there's a separate "enhance detail" algorithm, which is supposed to enhance very small detail. This way you can control edge sharpness and detail enhancement separately.

Quote:
Originally Posted by Neo-XP View Post
Also, we need a noise remover in madVR.
I don't really like noise removal much. Artifact remove: Ok. So if with "noise" you mean compression artifacts, then I'm with you, it would be nice to have that. But developing such algorithms is very difficult. For now it would be best to use some AviSynth script to do this kind of stuff. It should be done on the video in the original resolution, anyway, so it should be acceptable, doing this before madVR gets the images.

Quote:
Originally Posted by Neo-XP View Post
And "xstr" so I can define it to 0, because it sharpens the edges only (like unsharpHQ) and it's not a good thing for a high definition look
Agreed. That's why madVR has already set xstr to 0 by default.

Quote:
Originally Posted by trip_let View Post
Yes, either 75% or 100%.

I like the anti-bloating but don't see much either way in the anti-ringing.

I don't really prefer them but it looks different enough to be useful. But I don't really care about downscaling anyway.

Again, not of much interest to me but in general and including here I'd take the option that's more different, so the strict (soft).
Thanks.

Quote:
Originally Posted by Magik Mark View Post
Black screen is back while using nnedi3. Here are the circumstances:

This happens right after video transcoding a video is finished and I'm still watching video in Kodi via dsplayer. I don't know if other players are affected as well

I need to exit to FSE to bring back the picture then exit the player then play it again
So if you don't do video transcoding there isn't any problem? I'm not sure, maybe video transcoding is using OpenCL, too? In that it might conflict with madVR. It shouldn't, but if the drivers are bad, it could. Probably nothing I can do.

Quote:
Originally Posted by 70MM View Post
We want to operate A stretch, sharpness controls and blanking via an ipad
There's a network protocol, but it's relatively complicated.

Why do you need A stretch via ipad? madVR should be able to detect the correct aspect ratio of the movie and then should allow you to setup a profile for A stretch and one without, and you could let madVR automatically switch to the right profile. So everything could be fully automated.

Quote:
Originally Posted by 70MM View Post
Ive noticed if I set debanding to anything higher than low I get a lot of dropped frames.
Try activating the "don't rerender frames when fade in/out is detected" option. This option may require bigger than default queue sizes to work without frame drops, especially if you've configured the madVR settings in such a way that you GPU has to do some serious workload.

Quote:
Originally Posted by Betroz View Post
A friend of mine has an old Windows 7 machine, Core2Quad QX9770 @ 4000 and a GTX 980 @ 1520 MHz (nvidia 359.09 driver I believe) and a 1080p DLP projector. He uses the old MadVr 0.87.13 build with these settings :

This works on all the content he views with zero frame skipping. Now, if I use the same MadVr settings (and version of MadVr) on my system I get dropped frames! I have a Windows 10 x64 PRO system with Intel Skylake 6600K @ 4500 and GTX 980Ti @ 1300 (tested with 355.98 and 364.51 drivers) and my 1080p Panasonic VT30. Why is that? Is Windows 10 that broken? nVidia drivers? I do have a more powerful system than he...so this makes no sense to me.

What seems to work so far with my system is the newest MadVr 0.90.17 with these settings :

- Deinterlacing : OFF
- artifact removal : medium/high
- image enhancements : None
- chroma upscaling : NNEDI3 256
- image downscaling : SSIM 2D100%, AR relaxed, LL
- image doubling : Allways, NNEDI3 64 on ONLY double and quad LUMA
- image upscaling : Jinc AR
- upscaling refinement : CE1, ED1, LS 0.65, AR
- D3D11 Fullscreen exclusive mode
- dithering : Error Diffusion 2
- smooth motion off (Reclock)
- trade settings : all OFF
The latest madVR build may consume some more power than older builds, depending on which settings you're using. I'm always working on the algorithms, trying to improve the quality, so sometimes performance goes down a bit.

Generally, I'd recommend to spend your GPU resources where it matters most. E.g. as the other users already suggested, chroma upscaling is not really critical, so saving some performance in that area should allow you to spend your resources more effectively elsewhere.

Quote:
Originally Posted by Backflash View Post
Antibloat filter - very good! Personally i find 50% setting best for me.

Makes picture "foggy-ish" with superres+antibloat enabled, not sure why would anyone use it there.
I think most difference makes superxbr+antibloat in doubling.

Basically with my settings 720p anime became BR level on 1440p monitor - magic.
It's a bit more complicated with films, still looks better.

I also switched to Crispen edges 3.0 in IE, i find it much more usable now. I do not use settings that have antibloat in IE, something conflicts and makes picture weird.
Sorry can't explain better, 100% colorblind(meaning only black,white and billions of grey - seriously) so take all that with a grain of salt.
Thanks!

Quote:
Originally Posted by Uoppi View Post
After upgrading to 90.17 (from 90.15, I believe) and without changing any settings, render queues are suddenly struggling to fill up with some of my profiles.

Any ideas which new features might have become more taxing to the system compared to 90.15? I'd really prefer not having to go through all the profiles one by one again and try to guess what's different. FWIW, I'm not using any "gimmicks" besides doubling other than super-res x1.

EDIT: The only thing I can think of causing this is the tweaked upscaling/downscaling anti-ringing filters = much more performance hungry than before.
Anti-ringing for sharpening + Super-Res got slower.

Quote:
Originally Posted by XMonarchY View Post
Can the actual rips influence performance? My rig does Image Doubling - Luma with NNEDI3 32n SuperSampling always enabled on most 1080p 23-24p BD's and rips I made, but some specific rips stutter heavily with SuperSaming enabled with identical 23-24p 1080p content using identical rip settings and H.264 encoder settings... I obviously use the same madVR settings for both rip types - the ones that stutter with SS and the ones that do not.

- Any ideas why the above happens?
- Does madVR waste resources on rendering the black bars?
If you have frame drops / stuttering, check the Ctrl+J OSD to see which queue (from top to bottom) is the first one getting into trouble. That's your bottleneck. Could be the decoder, or the GPU, or something else.

madVR skips black bars if you enable black bar detection and activate "crop black bars".

Quote:
Originally Posted by 70MM View Post
Ok I have the sharpness and anamorphic stretch now working from my ipad but the notifications pop up on the top left hand corner for the profiles created. Is there a way to turn these OFF please, I don't want them to show?

As we have a movie club I don't want to inflict these notifications to my audience.....
Not possible atm. Will be possible in some future version. But it will take some time.

Quote:
Originally Posted by Werewolfy View Post
Yes, anti-ringing makes a difference. It's not a huge difference but it avoids some ringing with SuperRes so I think you should keep this option.

I didn't find an example where the anti-bloating filter makes a difference vs simply a lower SSIM value.

I didn't find an example where I found a difference regarding sharpness between these two options so I would choose the strict option to avoid ringing as much as possible. But I use a UHD TV so downscaling is not very easy to test for me, I don't have a lot of samples
Thanks!

Quote:
Originally Posted by austonrush View Post
Hey Madashi, I'm getting a blank screen maybe 10% of the time when switching to fullscreen exclusive mode. The player does not crash but, it does freeze my whole computer for about 20- 30 seconds then suddenly it starts working. I've only noticed it since 90.16 was released.

I'm using the latest nvidia driver on windows 10 64bit, mpc-hc, lav and reclock

Anyone else run into this issue?
Is it possible that your TV is having trouble synchronizing to an eventually modified refresh rate (or bitdepth)? Do you have 10bit output enabled?

Quote:
Originally Posted by austonrush View Post
- I do find antibloat useful for Adaptive sharpen 0.7 + crispen edges 1.0. It helps remove the paintbrush effect. I've settled on a strength of 125 for upscaling 720p content

- I've experimented with this on and off and if there is a benefit I'm having a hard time picking it out. I use super res 2 for 720p content anything higher is to aggressive

- YES but only when SSIM is at a strength of 100%. Otherwise you lose to much detail. Currently using at a strength of 100%. Anything higher and it cancels out the benefit of SSIM, IMO

- Relaxed. I have a panasonic VT series plasma and some are reporting they can see no difference. I can see a noticeable difference and relaxed retains more detail. strict is to aggressive. Relaxed does a good job of limiting ringing while retaining detail.
Thanks.
madshi is offline   Reply With Quote
Old 9th May 2016, 13:28   #37820  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by cxrn View Post
It's a real shame 'move subtitles' breaks with italic text - one line with just a "<i>" is enough to break all lines for the remainder of the movie.
It's on my to do list. But fixing stuff in XySubFilter is very painful for me, so I'm not sure when I'll get to that...

Quote:
Originally Posted by videonerd View Post
Hi madshi is DontRenderAfterStop still applicable in the latest build?
The code is still in there, so I suppose so.

Quote:
Originally Posted by Sunspark View Post
Found a performance bug in my setup with a fix. I was wondering why the rendering time on chroma was higher than it used to be in the past. It might be related to LAV 0.68.

Some numbers using a high resolution video playing in chroma upscaling mode only.

In LAV Video settings-

DXVA2 Copyback 1.6 ms
DXVA2 Native 2.6 ms

Yes, native was 1 ms slower than copyback.. which makes no sense to me, but that's what the HUD was saying.

This is with the following checked ON in madvr options:
use a separate device for presentation
use a separate device for dxva processing

Now, if you uncheck these options the rendering time in copyback jumps to 12+ ms.

I have no idea what interaction is at play here but the rendering time difference is huge with those two checked and unchecked.
Not sure why that happens, probably a GPU driver issue. In theory using separate devices should not help in any way, but it does.

Quote:
Originally Posted by Queiroz View Post
What should i do to remove the green artifacts when playing the file below?
Probably a GPU driver problem... Some NVidia drivers/GPUs can't handle NNEDI3 correctly.

Quote:
Originally Posted by Manni View Post
Sorry it took me a while to do more tests regarding the crash on first play, but I've been very busy with work.
I finally had a chance to spend a few hours trying to debug this, and here are the results.
My current config:
Win 8.1 Pro x64 with MCE
ADM 7870 With Crimson 16.4.1 (same issue with all Crimson drivers)
Latest MPC-BE x64 nightly, MadVR and LAV x64 beta

Problem: player crashes at launch (MPB-BE has stopped working) on first playback after a reboot. Playing the same file right after the initial crash (or any other file) works fine.
Once the first crash has occurred, playing the same file (or any other file) with a different player works fine.
There is no crash if EVR or EVR-CP is selected instead of MadVR.

I've tried the following:

Revert to Catalyst 14.12 (last stable catalyst for my GPU): same issue (rules out Crimson)
MPC-HC x64 instead or MPC-BE: same issue (rules out MPC-BE)
MPC-BE x86 instead of MPC-BE x64: same issue (rules out x64)
Using FSE mode: same issue (rules out non FSE mode)
Disabling the use D3D11 path: same issue (rules out D3D11 path)
Playing with the general settings in the rendering options (delay playback, disable desktop composition, use a separate device for presentation): same issue
Trying to launch MPC-BE and MPC-HC directly, without using MyMovies to select the film: same issue (rules out MyMovies)

I'm running out of idea, but it looks like this clearly establishes that the issue lies with MadVR as it's not the GPU driver, nor the player, nor MyMovies.

I'm going to try to go back to older MadVR versions, as this is not a problem I used to have, but I had already tried to go back as far as 0.89.x and the problem was still there, so not sure how much further I should go. What is the oldest version of MadVR that will still work with the current option/registry and won't mess up my configuration?

In case it matters, I was always trying to play the same film, a bluray at 23p (BDMV folder), but I have the same issue with any movie (well, at least BD folder) so it's not movie specific.
Finding the cause of crashes is always hard. Your tests seem to indicate that madVR is likely to be at fault. I'm not 100% convinced yet, but it's certainly very possible. The most problematic thing is that crashes that only seem to affect one user (or very few users), but not all users, are hard to pinpoint. Why does it only occur for those few users and not for the others?

It could be some 3rd party software conflicting. E.g. I've received tons of reports of Asus and other 3rd party software producing crashes in madVR. Could be something like that, just with the crash occurring outside of madVR, so that madVR doesn't catch and report the crash.

Anyway, how to solve the issue? I've really no clue right now. You could go back to older madVR builds to try to find whether one specific build introduced the problem. Of course going back means you're losing features. If you want to go this route, I'd suggest to do large steps, e.g. don't go from v0.90.16 to v0.90.15, then v0.90.14. Instead go from v0.90.16 to v0.85.0, then to v0.80.0. Or jump directly to v0.70.0 and if that works try v0.75.0 etc. Using large steps will make testing go much faster.

The settings are stored upward and downward compatible. But to be extra safe, you should store the "settings.bin" file, or the registry settings value if there's no settings.bin file. Then later you can restore the old settings by first double clicking "restore default settings.bat" and afterwards copying in the stored "settings.bin" (or registry value).

Quote:
Originally Posted by har3inger View Post
I gotta agree with you. At a glance, AB makes things closer to what they should look like, especially for the windmill bricks. Also, at least compared to bicubic 150, even with AB, SSIM is sharper. However, for the food pic, it's little harder to say that SSIM without AB is absolutely ruining the image. If I pixel peep, the table cloth pattern tends to jump too far forward, but at a distance, no AB looks better.

I think it depends a lot on view distance to decide if SSIM AB is better. Without AB, you definitely spoof a higher-res image with sharper/higher frequency features, and the inaccuracy isn't as bad or apparent at longer distances. This sharpening effect is probably better/more accurate than anything you could accomplish with adaptivesharpen or the likes. I'd say keep AB as an option for the user to decide on for SSIM (maybe labeled as "accurate" vs "sharp")
Quote:
Originally Posted by Warner306 View Post
I agree that anti-bloating is effective, but I'm not sure if removing the sharper SSIM is good or bad. Bicubic150 is not even close to as sharp. SSIM without AB might be preferable to users who want a really sharp image. Still, the less unnecessary options, the better. Most users will tick every checkbox available thinking it makes things look better.
Quote:
Originally Posted by Werewolfy View Post
I agree, SSIM 100% is better with Anti-Bloating enabled but is it better than just using SSIM 50% or 25%?
Quote:
Originally Posted by omarank View Post
Thanks for posting this comparison. Well, you did make me change my mind. After seeing these screenshots comparison, I did some more testing and found that anti-bloating at 25% strength actually helps when used with SSIM at a higher strength like 75% or 100%.
Thanks for your feedback. Maybe I'll tweak the anti-bloating algorithm *everywhere* to internally increase sharpening strength. Currently anti-bloating both changes the look of the image (less bloated) but also decreases the apparent sharpness. Would be nice to tweak it so that the apparent sharpness isn't changed, but only the look. But I'm not sure yet if I'll do that...

Quote:
Originally Posted by James Freeman View Post
Madshi, what rendering intent madVR uses by default when "this display is already calibrated" is selected?
Looks like a simple colorimetric intent that clips the out of gamut colors, right?
Atm simple clipping.

Quote:
Originally Posted by arrgh View Post
I have the same Problem with mvc-mkv...I tried xysubfilter but also the MPC-BE internal filters. Both times, with "Stereoskopic Subtitles disabled" and enabled as "side by side" I get the Subs in the Display plane...

the Subs are stored in the mkv as DVD-sub.
There seems not to be an Option (in MPC-BE, madVR or xysubfilter) to adjust the depth of the Subs...

I would be ok with an user-adjusted Option...I'm not talking about the "original Settings in the mpls"...
All MVC encodings I've seen have at least "some" subtitle depth information in them, and madVR uses that. Are we talking about MVC videos or something else (like user encoded side-by-side)?

Quote:
Originally Posted by cvrkuth View Post
When I watch a 24.000 fps movie with "use Direct3D 11 for presentation" enabled render queue and present queue are not full. Present time is unusually high too

With "use Direct3D 11 for presentation" disabled stats is ok.

Can I set a profile auto select rules for 24 fps movies?
This is a known problem with some GPU drivers, when using D3D11 with either 10bit output or "unusual" frame rates. Probably nothing I can do about. Usually reducing the number of prepresented frames helps.

Using profiles should work, I suppose. See first page of this thread.

Quote:
Originally Posted by James Freeman View Post
I noticed that the HDR to SDR algorithm in madVR saturates the color way beyond normal, I'll explain.
A standard 709 image converted to BT.2020 ST.2084 (ICC profile made by ArgyllCMS) exported as AVI and encoded with HEVC (primaries and matrix 2020, ST.2084 EOTF, P3 master display, 10,000 peak).

When playing back this video in madVR the image should look identical to the original 709 image right?
The resulting image in madVR is dark and more saturated compared to simply reversing the 2020 st.2084 icc profile (back to linear).
That lets me believe that something is wrong.
Every time in the recent past, when a user sent a user-encoded sample to me which indicated that madVR might do something wrong with the colors, after spending hours on debugging I found that the sample was encoded incorrectly, either due to user error or bugs in other software.

Because of that at this point I'm not eager to look into user-encoded samples, especially as long as madVR is the only renderer supporting HDR, so we have nothing to compare to. Official samples only at this point, sorry.

Quote:
Originally Posted by James Freeman View Post
Apparently madVR algorithm is much more complex with range conversions much smarter than I am.
It looks like it is not a simple curve as I assumed, it is dynamic gamut and average picture level dependent.
Madshi, can you just explain how it works, I will be grateful.
As I mentioned before, the algorithm is not in its final state atm, so I don't think it makes much sense for me to explain how it works now, when it will change "soon".

Quote:
Originally Posted by RenderGuy2 View Post
Would it be possible for MadVR to support:

MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
MEDIASUBTYPE_XYZC {435A5958-0000-0010-8000-00AA00389B71}
FORMAT_VIDEOINFO2 {F72A76A0-EB0A-11D0-ACE4-0000C0CC16BA}

This is a 12-bit XYZ format. There are a couple of relatively inexpensive DirectShow filters that can decode jpeg2000 pretty quickly and can output 12bit xyz (without forcing a conversion to rec709). The conversion to display color can easily be done by the user with either a 3DLUT or pixel shaders (linearize, 3x3 matrix, gamma), or a combination of both.

I have no idea if it is simple or difficult to add support for another pixel type. Thank you.
It wouldn't make sense for me to try to implement something like this "blind". So basically I'd need to be able to play some sample files directly on my PC, preferably without having to spend any money.

Quote:
Originally Posted by Firebird View Post
Hey madshi, thanks for madVR! Do you mind feature requests? I have a video which was IVTC'd (not by me), but it still has duplicate frames. When i force film mode on it it doesn't seem to work well. Tdecimate works though. It's not a deal breaker, but it's annoying. Can you fix this? i'll send you video sample if you need it.
Can you send me a sample which shows this? I'm not sure when I'll find time to look into it, but samples where IVTC fails to work as expected are always welcome.

Quote:
Originally Posted by James Freeman View Post
madshi can you please add DCI-P3 D65 to madVR?
Add it where and how and why? It's already supported if SMPTE 2086 metadata specifies it.

Quote:
Originally Posted by zaemon View Post
sometimes madVR does not switch to 3D stereoscopic mode and resolution when I play back an MVC video. That's a problem because I'm outputting at 4k and thus the video is played back in 2D at 4k. Maybe having some kind of 1080p23@3D display mode would do the trick. Right now I'm using two profiles, one with 4k resolutions and one with 1080p. Using 1080p display mode always ensure that 3D stereoscopic mode is activated. Using 4k display mode don't.
In theory madVR should manually switch to 1080p and only afterwards try to enable 3D, exactly to solve the issue you're reporting. Not sure why that doesn't seem to work for you.

Quote:
Originally Posted by James Freeman View Post
madshi, the 10,000 nit HDR setting will not respect my selection and clip the highlights.
It also exhibits random behavior when pausing the video or resizing the window.
There is some unstable interaction between the Metadata and the selected nit peak setting.
When I select 10,000 nit I need to see all the 64-940 range without madVR clipping, stretching, or compressing anything, even if above 1000 nit input (metadata peak, step 721) there is no content.

Happens with the official HDR demo videos or my videos.
Are there any official samples with 10,000 nit? I've only seen max 4,000 nits samples.

Quote:
Originally Posted by sat4all View Post
Any idea why madvr dxva image upscaling is clipping btb and wtw?
madVR is trying to convince the GPU/GPU drivers to not clip BTB/WTW, but the GPU drivers don't always listen.

Quote:
Originally Posted by oddball View Post
Since a few days ago MadVR is not working in FSE mode under Windows 10. I get a blank screen on a secondary or tertiary display. If I right click on XYsubfilter on the system tray the image comes back. If I bounce between FSE on secondary/tertiary a couple of times MPC-HC crashes. If I try skip frames it locks up. It works using Windows overlay or DX11. I tried different (older) versions of MadVR and MPC-HC plus older AMD drivers (default Win10 version).
Well, if this is "since a few days ago" something must have changed. Not sure what, though. It's really impossible to diagnose from here.

Quote:
Originally Posted by James Freeman View Post
Can you please inform me what LAV and madVR does with chroma location, how do they decode and display it.
Is madVR "fixing" or moving the chroma position to a different place to re-align the 0.5 pixel shift?
I see that I have to use at least Billinear scaling for the chroma to display correctly.
Probably LAV reports the correct placement, but madVR currently ignores it and always applies the pixel shift, because anything newer than MPEG1 uses that.

Quote:
Originally Posted by Sunset1982 View Post
Does MadVR already support DolbyVision HDR? Will it support it in future?
Dolby Vision specs are not available to me.

Quote:
Originally Posted by VAMET View Post
I have got 3D effect, but English is not my first language, so I use Polish subtitles. I think there is something wrong with subtitles shown on my screen, they are blurred/parted, so my eyes hurts, when I watch such movie. Are there any options or external filters, which I should use instead of those built in MPC-HC?
Which subtitle renderer are you using? Try the MPC-HC internal one and XySubFilter. Do you see the same problem with both?

Quote:
Originally Posted by cxrn View Post
I heard madshi planned on making a "add black bars" feature, but I'm not sure if it's still is on the table or not.
I don't remember. What was this supposed to do?

Quote:
Originally Posted by Nabukodonosor View Post
I have a problem with this great renderer. I breaks a lot. I've attached my bug report. Please help. Thanks
You're using a very old madVR build. Try the latest one. If that still crashes, send a bug report to me.
madshi is offline   Reply With Quote
Reply

Tags
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 12:43.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.