Media Player .NET (MPDN) - D3D HQ GPU Video Renderer [v2.49.0/v1.31.0 27 Dec 2018] [Archive] - Page 81

ryrynz

8th March 2016, 23:38

Lav 0.68 has been released. I've made an inquiry with Alex about possibly releasing a new version of Sanear since it's been a while.

Belphemur, could you fix the updater so it always checks for the latest version before displaying the update window rather than relying on the cache? I believe you're already aware of this happening.

For example, I was prompted to update to 2.46 and 1.26.1 so I shutdown the player, afterwards
it refreshed the extensions to 1.26.2, ideally the notification should be up to date.

Anima123

9th March 2016, 08:15

I have found that video source with good quality and resolution, for example good encoded 720p source, tend to be overly-sharped, if using image-doubler, e.g., Super-xBR with default value together with SSIM. In this case, adjust the 'sharpness' down to a lower value, like 0.40 will give visually pleasant quality.

As contrary, if the source is low-resolution, like 720 x 404, Super-xBR with low 'sharpness' most likely will blur the image, in this case 'sharpness' as high as 80-100 will result in visually pleasant images.

I am not sure anyone has experienced the same. The reason behind it might be that lower resolution will need more "sharpness" to get closer to the 'grand truth' images, while images with higher-resolution and high quality just need to keep the sharpness before it has been overly sharpened, and hence brings other kind of artefacts.

What's your opinion on this?

Zachs

9th March 2016, 11:10

Kinda makes sense I suppose. You should try using the conditional script to make sure you have the sharpness you like for each resolution or a range of resolutions. Perhaps share it here too so others could use it.

aufkrawall

9th March 2016, 22:19

Is it by design that sometimes there are delayed frames reported with Fluid Motion?

Zachs

9th March 2016, 22:34

No. It only happens when the driver or GPU missed the presentation window. Happens more often if your display rate is higher because the window is smaller.

YxP

17th March 2016, 14:06

Any chance of an update for 3DLUT script behaviour? Preferably something like MVR's, where you can have more than one screen available and the player takes care of disabling OS's calibration. Well, you probably know what I mean. I have two calibrated screens and using 3DLUT right now is way too cumbersome.

aufkrawall

17th March 2016, 14:10

No. It only happens when the driver or GPU missed the presentation window. Happens more often if your display rate is higher because the window is smaller.
Do you think this is visible? :)

huhn

17th March 2016, 14:16

Do you think this is visible? :)

if the "wrong" frame is repeated is sure is.

Zachs

17th March 2016, 21:50

No. Not when your refresh rate is sufficiently high.

YxP

18th March 2016, 14:09

The player crashes every time I try to open a file if I have selected a script group to be used (using the premades by aufkrawall). I can, however, select one during playback and it works. Using latest LAV nightly -10. Win 8.1 x64.

Blackfyre

18th March 2016, 19:02

Hi Zacks,

Hope all is going well, I haven't been active for a long time here.

I'm trying to watch IP Man 3, and I have xySubFilter chosen and it's not loading the subtitles into the player. It's working fine with MPC but with MPDN even the tray-icon for xySubFilter is not showing. Maybe I'm missing something simple. Any help or guide regarding setting up xySubFilter with MPDN or just choosing it via options under subtitles should be enough?

Running the latest version + latest extensions.

I might send you a private message with a new guide/article for SVP 4 Full Version to add to the OP in a week or two. As the settings and their effects have changed significantly since SVP 3.

Regards, Blackfyre.

Shiandow

19th March 2016, 01:33

Blackfyre

19th March 2016, 10:31

Hi all,

I've manage to improve the reconstructive bilateral scaler enough that I consider it (probably) ready for general use. I would really appreciate if you could try it out and see if there are indeed no more unacceptable artefacts.

Personally I'm pretty happy with the way it looks, in some cases it really manage to recover quite a bit of the chroma information, which is a lot more noticeable than you'd think (even at native resolution, but especially when upscaling).

We'll do some testing later today or tomorrow, thanks for the update Shiandow.

The player crashes every time I try to open a file if I have selected a script group to be used (using the premades by aufkrawall). I can, however, select one during playback and it works. Using latest LAV nightly -10. Win 8.1 x64.

I can confirm this, the script group examples linked in the OP are crashing MPDN when used.

Using Windows 10 latest built here.

MPDN x86

ryrynz

19th March 2016, 11:11

Shiandow

19th March 2016, 18:34

Being guided by luma is appaears to making the result appear to be more accurate, but it's super sharp at 0.5.. even at 0.1 it's sharper than madVR's Bilateral.

Using madshi's chroma test video from the movie Samsara (Two guys in red robes blowing horns) it really doesn't look that good even at 0.1 (at least with that example)
Although that is one of the worst examples you can use to show bilateral's faults.

Doesn't seem to be stable either, MPDN stops working when using it, no dump. I have to switch to it from another scaler, I can't load a file with it activated.

Still prefer super-xbr for chroma, you should add it into MPDN, perhaps it too could be improved by guiding by luma? If that's possible..
Would be interesting to see how Hyllian's recent changes affecting ringing affect it vs madVR's older version.

I think I know a way to make it softer, which would probably be better for more noisy sources as well. The 'strength' setting ideally shouldn't affect sharpness, it only affects how far the algorithm is willing to go to reconstruct the chroma. Not sure if the 'reasonable' values are spaced well though, could be that you'll get reasonable results at 0.01, or even 0.001 (but not 0). Do you have a link to the chroma test video? That would probably help.

And I think it's doable to make Super-Xbr scale chroma, but I don't see an obvious way to make it use the high resolution luma to guide the interpolation. Maybe Hyllian can find a way, but maybe not.

I'm also not sure what's happening with the stability issues, I'm not seeing any at the moment. Perhaps I inadvertently fixed it, but that seems unlikely.

Anima123

19th March 2016, 23:30

For natural videos, the bilateral chroma upscaling dose not seem right if 'strength' is set to 0.10. Since 0.10 is the step, it does not look good under 0.20 as far as I can figure.

Shiandow

19th March 2016, 23:46

In what sense doesn't it look 'right'?

Anima123

20th March 2016, 00:24

Some kind of 'vague' at the edge of human faces. I am not very sensitive to chroma than luma, that I cannot describe it more precisely.

Shiandow

20th March 2016, 00:29

Some kind of 'vague' at the edge of human faces. I am not very sensitive to chroma than luma, that I cannot describe it more precisely.

Haven't noticed that before, I'll see if I can find out what's happening. Some screenshots would help though.

Anima123

20th March 2016, 01:52

Here's 2 screenshots:

strength = 0.10
https://www.dropbox.com/s/qwfnijzour2jegv/Img01_bilateral_0.10.png?dl=0

strength = 0.20
https://www.dropbox.com/s/hztmghcidmmu12t/Img01_bilateral_0.20.png?dl=0

It's not as obvious as I thought, but 0.10 seems to have a little more blurry.

Shiandow

20th March 2016, 02:35

Hmm, seems the 'strength' parameter isn't too useful then. Unless anyone has found a situation where a value other than 0.5 looks better? If not then I'll simply remove that parameter, no use keeping an option that doesn't ever improve the image.

So if anyone finds a case where strength = 0.5 doesn't look good, or some other value of strength looks better, I'd love to hear about it.

Anima123

20th March 2016, 03:32

I would prefer strength = 0.2 ~ 0.3, if I have to choose.

ryrynz

20th March 2016, 06:01

Here's 2 screenshots:

It's not as obvious as I thought, but 0.10 seems to have a little more blurry.

Just like in madVR, it's the reds that really show the difference, that's not a good shot to use for comparison IMO..

Hmm, seems the 'strength' parameter isn't too useful then. Unless anyone has found a situation where a value other than 0.5 looks better? If not then I'll simply remove that parameter, no use keeping an option that doesn't ever improve the image.

So if anyone finds a case where strength = 0.5 doesn't look good, or some other value of strength looks better, I'd love to hear about it.

0.5 was incredibly sharp.. Too high IMO. Test on these videos 1 (https://mega.nz/#!DRZzyAZY!Oq-RxDtp9K0hRbmsa8CQZXqFwy_sZDoe5JdNLjaRM4M), 2 (https://mega.nz/#!jcAlzC5K!oFGoQyhiiHfFq1PPi3rfYbw7r7UlKIXOdhp-9igKJ18) taken from the madVR thread. I use the first frame from both. Focus on the robes of the first video and the Japanese lettering in the speech bubble on the second.

I personally find super-xbr 100 in MadVR to look the most accurate or natural to my expectations.

Comparison using the second video super-xbr 100 vs bilateral @ 0.1
http://screenshotcomparison.com/comparison/166076

Comparison using the first video super-xbr 100 vs bilateral @ 0.1
http://screenshotcomparison.com/comparison/166078

Shiandow

20th March 2016, 16:55

In both cases the chroma limiter seems to be having some issues, do they look better if you turn "Improve Chroma Reconstruction (Ctrl+Shift+C)" off ?

Shiandow

21st March 2016, 20:42

Okay, after some more testing I've concluded that the reconstructive method still has some more potential, however to push it to it's limit I need a better way to filter out artefacts. For now the current version is my best attempt at balancing accuracy vs stability.

Anima123

21st March 2016, 21:33

Keep up the good work Shiandow.

And I did found that "Improve Chroma Reconstruction" tend to destroy the whole image in some cases, so I usually set it as off.

Shiandow

21st March 2016, 22:06

Yeah, I might need to revise the 'improve chroma reconstruction' at some point. It doesn't handle some edge cases well. Fixing it is a bit tricky though (GPUs don't like situations with lots of cases that need to be handled differently).

ryrynz

22nd March 2016, 10:43

In both cases the chroma limiter seems to be having some issues, do they look better if you turn "Improve Chroma Reconstruction (Ctrl+Shift+C)" off ?

Yes. When disabling ICR, the speech bubble looks better than even super-xbr ( got similar results using madVR)
It was also the option responsible for darkening various elements on the robes. My guess is that this could perhaps be toned down a little perhaps.
Of course the robes edges are still all blown out even with ICR disabled.

It will be interesting to see how madVR's Super-xbr handles chroma with two levels of anti-ringing. ATM I'm no fan of enabling it when using it on chroma, but that might change with the new release.
A luma guided ICR enabled double anti-ringing super-xbr might just end up being top tier chroma upscaling :D

ryrynz

27th March 2016, 11:41

Not sure if due to recent changes but using Fluid Motion is like watching 30+ FPS content from just 24 FPS.

There's an almost Frame Creation type of effect from it producing a pop effect almost 3D like in appearance.
Zach has talked about MPDN's benefits having been built from the ground up to attain the very best frame presentation but this is the first time I can say I've ever seen it in practice.

Shiandow

27th March 2016, 19:44

Hi,

I've made some improvements to the reconstructive bilateral method again, I've managed to detect when there's not enough luma information to reconstruct the chroma, and made it avoid those situations. It's not perfect, but it never will be, and I think any further (significant) improvements will require a different approach.

So, I'd be interested to know what you all think. Is it strong enough when it can be? Or is it still too strong where it shouldn't be? Does it work well with SuperChromaRes, or is SuperChromaRes not even necessary anymore?

Also, if you want a quick comparison with other chroma scalers, here (http://i.imgur.com/qFUivrc.png) is the result for a popular benchmark image, and here (http://screenshotcomparison.com/comparison/166640) is the same image when using MadVR's soft-reconstructive and superXbr chroma scaling algorithms. For this image at least the reconstructive bilateral method is clearly superior.

Cheers,
Shiandow

Shiandow

28th March 2016, 16:51

madshi

28th March 2016, 17:03

That does look pretty promising!

Anima123

28th March 2016, 21:37

Here are some more comparisons of the reconstructive bilateral method vs simple bicubic scaling:

Netflix logo (http://screenshotcomparison.com/comparison/167142)
Samsara sample (http://screenshotcomparison.com/comparison/167140)
Anime sample (http://screenshotcomparison.com/comparison/167141)

Isn't Bilateral a little bit too strong for Samsara sample? I also noticed more alias on the red robe of the lama.

Does smaller strength work smoother here?

Shiandow

28th March 2016, 22:34

Lowering the strength to 0.2 makes it smoother, it pretty much reverts to bicubic on anything except edges, but whether this is 'better' is up for debate. Also not sure how well such a low strength performs on other images.

There's not much that can be done to fix the aliasing though, the aliasing is pretty clearly present in the luma channel, so the method assumes it must therefore be present in the chroma channel as well (which, for all I know, might have been the case).

madshi

29th March 2016, 09:05

Yes, Samsara has clearly visible aliasing in the luma channel. Which means any luma guided chroma upsampling algorithm will run into some trouble with that sample. But FWIW, the new Bilateral seems to have less trouble with that than most other luma guided algorithms I've seen so far, which is one reason why I have hopes for the new Bilateral algo to become the new "best" chroma upscaler. But we'll see, we'll need some wider testing from users with lots of different content.

aufkrawall

29th March 2016, 13:47

This sample does not just only show nasty banding, also the red arrows are quite critical for chroma upscaling:
http://www36.zippyshare.com/v/LZSWS3My/file.html
Shiandow, you might want to take a look at it again with new bileteral scaler? :)

Shiandow

29th March 2016, 14:36

It seems to perform quite well on that image (comparison (http://screenshotcomparison.com/comparison/167276), scaled to 1080p), did the previous version have some trouble with that image? Only difficult part is the gradient near at the tip of the arrow, it's hard to be sure what that was supposed to look like, but the result looks pretty decent.

The banding is horrible though, and possibly unrecoverable (that image not only has horrible banding all over it, but also one weak intentional band running over Turkey, it's nearly impossible to remove one but not the other).

If you really want to give the bilateral algorithm a run for its money you should try images with faint fine details in the luma channel, that can really trip it up quite badly (for obvious reasons).

madshi

29th March 2016, 16:18

The chroma position seems to be shifted to the right in that sample, for some reason.

Shiandow

29th March 2016, 16:38

I think it's supposed to look that way, the curve matches up nicely, and if the chroma offset were incorrect then bilateral should have fixed that to some extent (so you'd see the chroma jump sideways compared to bicubic).

madshi

29th March 2016, 16:57

The reason I think the chroma position is shifted is that the red chroma of the arrow-head bleeds into the green background on the right side of the arrow-head. If the chroma were positioned correctly, this bleed should not occur, or at least it should occur everywhere, e.g. on the top & bottom of the curve, not just on the right side of the arrow head.

huhn

29th March 2016, 17:03

the rest of the images doesn't show the same shift.

i think is was moving and maybe even intentional

or there was so little bit rate that the chroma couldn't be changed.

this would not be the first time i would see a part of chroma from an old frame in broadcast.

madshi

29th March 2016, 17:10

You're right, it's probably a compression problem. Although it's the same throughout the whole time the arrow moves. However, when the arrow is in its final position, chroma suddenly becomes nicely centered everywhere. So at least the chroma position is not generally shifted in the whole video. Well, anyway...

Shiandow

29th March 2016, 17:13

Ah, that might explain why you don't see a chroma shift in the curve, that part doesn't move. It also explains why the algorithm has some problems with the arrow head (since luma and chroma don't match up).

huhn

29th March 2016, 17:45

to be honest your algorithm shouldn't try to fix this.

i mean how should sub sampling and compress create such a huge gap there?

on a good source it shouldn't touch more than the neighbor pixel or something like that.

Shiandow

29th March 2016, 19:26

on a good source it shouldn't touch more than the neighbor pixel or something like that.

Agreed, but I think we can agree that that particular image wasn't a good source. Also when you scale subsampled 720p source to 1080p, the chroma pixels become huge.

huhn

29th March 2016, 19:46

you want to use bilateral for direct scaling not just for 4:2:0 to 4:4:4?

Shiandow

29th March 2016, 20:00

Haven't tried that yet, but I suppose it would work. Might even be better in some cases, although other than performance I don't think you'll notice much difference.

But even if you go 720p 4:2:0 -> 720p 4:4:4 -> 1080p, you make any problem the chroma signal has a lot bigger.

Hrxn

29th March 2016, 21:59

Hi guys,

I am new here, so please bear with me a bit :)

Okay, so a friend of mine recommended me MPDN as a great new media player, and now I'm curious and want to try it out.

I just read the project page (http://www.zachsaw.com/mpdn/), requirements are in line, but this is a bit strange to me:

Certificate
zachsaw.com Software Certificate - Install this first. You only need to do this once.

What is this about? I don't know, I think I've never seen this for an application before. Isn't this a security risk? Installing a certificate to the trusted root store??

Doesn't MPDN work without the cert?

ryrynz

29th March 2016, 22:23

What is this about? I don't know, I think I've never seen this for an application before. Isn't this a security risk? Installing a certificate to the trusted root store??

Doesn't MPDN work without the cert?

It works fine without the cert. Read below.

This prevents the annoying prompt that you're running an application from an unknown publisher for software from zachsaw.com.
It also gives you the added security of preventing you from getting desensitised by these prompts (many already are by simply pressing 'Yes'
whenever they're prompted). It also prevents other developers from claiming their software is from zachsaw.com - another security feature.