madVR - high quality video renderer (GPU assisted)

[aufkrawall]

This sounds like an issue which is out of my control. If the problem goes away as soon as an UI element is visible then the chance is rather low that it could be my fault, because (unless you're using FSE) madVR neither cares nor even knows whether an UI element is visible or not.

Do you have the same problem in FSE mode? Have you tried totally closing Afterburner (and any other GPU related tools)?

I tested again and it was still Afterburner. Situation was a bit different, this time FSE works without drops and it seems like it was important that Afterburner was never already started once during the Windows session.
So, FSE works for me. Hope it remains that way.

[huhn]

is it normal that super XBR can be really slow depending on the scene?

i'm talking about 50 % GPU to ~95% (dropping frames) reproduce able. the bus interface is spiking at the same time from 2-8 % to 22-60%.

and it looks like Vram usage is a lot higher too with that version or not stable. i'm not sure yet. even with a render queue of 4 i'm kind of capped.

what so ever i have huge troubles with 90.16.

[Werewolfy]

I don't really know why Reclock would effect frame drops, especially when using forced film mode. You could try create a debug log, then I can have a look for which exact reason the frames were dropped. Remind me: The OSD does list the frame drops, right? If it doesn't, the debug log won't help.

I don't know why the queues don't fill. That's a question you'd have to ask the decoder developer (not sure if it's nevcairiel in your case?). Maybe the DVD Navigator somehow limits this? I've no idea, I believe it's totally outside of my control.

Yes the OSD does list the frame drops. I will create a debug log this week end.

I use LAV Filters so yes I will ask nevcairiel.

[madshi]

is it normal that super XBR can be really slow depending on the scene?

i'm talking about 50 % GPU to ~95% (dropping frames) reproduce able. the bus interface is spiking at the same time from 2-8 % to 22-60%.

and it looks like Vram usage is a lot higher too with that version or not stable. i'm not sure yet. even with a render queue of 4 i'm kind of capped.

what so ever i have huge troubles with 90.16.
No that's not normal. I see no reason why super-xbr should depend on the scene. Are you sure it's not something else, like e.g. black bar detection? Although that would affect CPU, not GPU. If there really is a problem, I'll need more information. E.g. does it work if you restore default settings (other than super-xbr)? And if you go back to v0.90.15 the problem totally goes away?

Are you saying v0.90.16 is wasting GPU RAM, maybe even leaking it? How do v0.90.15 and v0.90.16 compare in that area?

[nevcairiel]

I use LAV Filters so yes I will ask nevcairiel.

LAV really has no influence on this. It just decodes frames it gets from the source and sends them onwards. DVD decoding is already optimized for latency because the DVD Navigator is known to bug out when the latency is too high - which is what might simply be happening here.
To further reduce latency, I would recommend to use software decoding in LAV Video for DVDs, its only SD MPEG-2, any recent CPU laughs at that, and it reduces latency a bit.

[madshi]

LAV really has no influence on this. It just decodes frames it gets from the source and sends them onwards. DVD decoding is already optimized for latency because the DVD Navigator is known to bug out when the latency is too high - which is what might simply be happening here.
So the LAV Splitter is not involved at all here? The DVD Navigator does all the work? Can you provide a bit more information on your experience with latency being a problem? Could that explain why Werewolfy has problems when letting Reclock change 25fps to 24fps?

[nevcairiel]

So the LAV Splitter is not involved at all here? The DVD Navigator does all the work? Can you provide a bit more information on your experience with latency being a problem? Could that explain why Werewolfy has problems when letting Reclock change 25fps to 24fps?

Yes, the DVD Navigator does all the work. And it is limited in how far it allows to "desync" the video/audio streams, ie. how many video frames you can pull out of it before it waits for audio to also be consumed.

In short, it won't let you queue up as many frames as you want, so the decode queue running low is not necessarily unexpected if the frames get drained instantly into all the other queues and you hit the limit from the DVD Navigator then.

[GCRaistlin]

Could you please download the PDB debug symbols for LAV and MPC-HC and create another freeze report with those "installed" (just need to be in the right folder)?

I've updated the freeze report: http://pastebin.com/LRxRi0Mu

[James Freeman]

Thank you for the new build.

Wow, the anti-bloating thing is great!
Basically a high-pass filter before the Sharpening algorithm right? Isn't something similar is already part of these sharpening algorithms, or you just let us play with the value inside the algorithm?
To me it looks much closer to how my eye would see detail in reality once I find the right % value.

With 1080p content I like AdaptiveSharpen at 1.0-1.5 with anti-bloating 75% to 100%.
It reveals the net under the actors wigs, skin pores, bad shave and much more fine detail we probably not intended to see.

[bcec]

some questions just so I make sure I understand these right:
1) What's the difference between processing > image enhancements, vs upscaling refinement? first applying non-upscaled image, and latter applying upscaled image? If both are set, two sets of image enhancements/refinement done?
2) For upscale refinements, is it preferred to refine after every 2x, or the final image?

cheers!

[Werewolfy]

LAV really has no influence on this. It just decodes frames it gets from the source and sends them onwards. DVD decoding is already optimized for latency because the DVD Navigator is known to bug out when the latency is too high - which is what might simply be happening here.
To further reduce latency, I would recommend to use software decoding in LAV Video for DVDs, its only SD MPEG-2, any recent CPU laughs at that, and it reduces latency a bit.

Thanks for responding before I had the time to ask :)

The problem is I already use software decoding and experience these issues... But you're right with hardware decoding it's worse.

Yes, the DVD Navigator does all the work. And it is limited in how far it allows to "desync" the video/audio streams, ie. how many video frames you can pull out of it before it waits for audio to also be consumed.

In short, it won't let you queue up as many frames as you want, so the decode queue running low is not necessarily unexpected if the frames get drained instantly into all the other queues and you hit the limit from the DVD Navigator then.

So now I know why it does that... But it seems that there's no solution :(

[nevcairiel]

So now I know why it does that... But it seems that there's no solution :(

In the test videos you once uploaded a while ago, you used extremely long queues (24 GPU, 15 present), which would of course contribute to this problem. I haven't really followed what your latest settings might have been, but alas for DVD the best course is to work with the smallest queues as possible in all areas to avoid tripping this limitation in the DVD Navigator.

[huhn]

No that's not normal. I see no reason why super-xbr should depend on the scene. Are you sure it's not something else, like e.g. black bar detection? Although that would affect CPU, not GPU. If there really is a problem, I'll need more information. E.g. does it work if you restore default settings (other than super-xbr)? And if you go back to v0.90.15 the problem totally goes away?

Are you saying v0.90.16 is wasting GPU RAM, maybe even leaking it? How do v0.90.15 and v0.90.16 compare in that area?

i'm looking into that. i simply wanted to know if the change to super xBR is taking extra resources.

90.15 was working. even with a render queue of 6.

looks like it is 3d LUT related. i don't get rendering spike without it. but a 3D LUT should always take the same processing power with CFR content.

i come back later with more information.

[Werewolfy]

In the test videos you once uploaded a while ago, you used extremely long queues (24 GPU, 15 present), which would of course contribute to this problem. I haven't really followed what your latest settings might have been, but alas for DVD the best course is to work with the smallest queues as possible in all areas to avoid tripping this limitation in the DVD Navigator.

I only used those long queues for my tests. The best settings I found was 23 CPU, 14 GPU and 6 present so I will stick with that for DVD.

Thanks for providing me those details.

[XMonarchY]

I am having a problem playing many video games' BK2 files (BIK files). I know Bink Smacker RAD Video Tools play them, but without all the benefits of madVR. Some do load in MPC-HC and play just fine, but most either don't load at all or load only the sound.

Is this:
- madVR issue/limitation?
- LAV Filters issue/limitation? I am using the very latest 0.68
- MPC-HC issue/limitation?

Is there anything I can do to make all the BK2 files work with MPC-HC, LAV Filters, ReClock, and madVR? I am sorry if this is the wrong thread to ask this. I don't know where the problem is...

[sneaker_ger]

Is this:
- madVR issue/limitation?
Set renderer to something else (e.g. EVR) and find out. It's not rocket science.

[nevcairiel]

Is there anything I can do to make all the BK2 files work with MPC-HC, LAV Filters, ReClock, and madVR? I am sorry if this is the wrong thread to ask this. I don't know where the problem is...

Bink Video 2 is not supported by LAV/FFmpeg, so if its a new-ish bink file, you are out of luck.

[clsid]

The Bink formats are simply just partially supported by FFmpeg (which LAV uses). Nothing you can do about it.

[aufkrawall]

I tested again and it was still Afterburner. Situation was a bit different, this time FSE works without drops and it seems like it was important that Afterburner was never already started once during the Windows session.
So, FSE works for me. Hope it remains that way.
As if it wasn't already confusing enough:
Even in FSE, I'm getting one dropped frame after 5 minutes of watching. So I disabled DXVA2 copyback in LAV to see if it makes any difference, and indeed: Now queues also get filled properly in windowed fullscreen and I can watch more than 5 minutes without any issues so far. However, still low level monitoring with driver must be used in Afterburner.
Hope they fix this at some point in the drivers with new GPUs (got the same problem with Sapphire Trixx OC tool instead of Afterburner) or else one might need to use an external GPU (e.g. of APU) for decoding if one wants to use voltage control or custom fan curve with HEVC 10 bit content.

[madshi]

Yes, the DVD Navigator does all the work. And it is limited in how far it allows to "desync" the video/audio streams, ie. how many video frames you can pull out of it before it waits for audio to also be consumed.

In short, it won't let you queue up as many frames as you want, so the decode queue running low is not necessarily unexpected if the frames get drained instantly into all the other queues and you hit the limit from the DVD Navigator then.
Ah, I see, thanks. And I suppose we could run into some sort of weird trouble if the "desync" gets too much for the navigator to handle, so it's probably best to try to avoid reaching that special navigator threshold?

I've updated the freeze report: http://pastebin.com/LRxRi0Mu
Ok. It seems the Microsoft DVD Navigator internally freezes when MPC-HC and madVR try to ask it for information at the same time from within different threads. This looks like a deadlock bug inside of the DVD Navigator.

Maybe it's time to upgrade to a decent media player OS? Windows XP is *really* outdated by now. I've not had any such freeze reports from anyone using Windows 7 or newer. I'd suggest to upgrade to Windows 8.1.

Or maybe there's a way to get a newer version of the DVD Navigator installed on your XP? E.g. according to the bug report you're still on SP3. Have you tried updating to SP6? Maybe a new DVD Navigator version is part of SP4-6? I've no idea, to be honest.

In any case, it seems out of my control. Short of not contacting the DVD Navigator *at all*, I don't see what I could do. And if I don't contact the DVD Navigator at all, several madVR features would stop working (e.g. a proper FSE seekbar for DVD playback).

Wow, the anti-bloating thing is great!
Basically a high-pass filter before the Sharpening algorithm right? Isn't something similar is already part of these sharpening algorithms, or you just let us play with the value inside the algorithm?
To me it looks much closer to how my eye would see detail in reality once I find the right % value.

With 1080p content I like AdaptiveSharpen at 1.0-1.5 with anti-bloating 75% to 100%.
It reveals the net under the actors wigs, skin pores, bad shave and much more fine detail we probably not intended to see.
I'm glad you like it. FWIW, if you like AdaptiveSharpen with 100% anti-bloating, have you tried "crispen edges" yet? It should look quite similar to that, without even needing the anti-bloating filter.

And no, LumaSharpen or AdaptiveSharpen don't have any such anti-bloating included, that's why I'm making it available externally.

first applying non-upscaled image, and latter applying upscaled image? If both are set, two sets of image enhancements/refinement done?
Correct.

For upscale refinements, is it preferred to refine after every 2x, or the final image?
There's only a difference if you enlarge much more than 2x. In that case refining after every 2x step might make more sense, because sharpening isn't as effective after one very large upscaling step, but of course sharpening after every 2x step is more work for the GPU, so you need a faster GPU, or you need to downgrade the algorithms you're using.

i'm looking into that. i simply wanted to know if the change to super xBR is taking extra resources.

90.15 was working. even with a render queue of 6.

looks like it is 3d LUT related. i don't get rendering spike without it. but a 3D LUT should always take the same processing power with CFR content.
It may vary with the scene because if the scene has "unusual" colors, that part of the 3D LUT might not be in the texture cache. A lot depends on how the GPU driver handles the 3D LUT, how much of it is in the texture cache etc. Will depend on GPU texture cache size, driver version etc. But the last time a user reported a problem like this was a LOOONG time ago. I had thought issues like this had been a thing of the past. In any case, there's probably not much I can do about it, except maybe allowing smaller 3D LUTs to be used.

[GCRaistlin]

In any case, it seems out of my control.

But madVR versions prior to 0.90.4 almost haven't had this issue...

[huhn]

It may vary with the scene because if the scene has "unusual" colors, that part of the 3D LUT might not be in the texture cache. A lot depends on how the GPU driver handles the 3D LUT, how much of it is in the texture cache etc. Will depend on GPU texture cache size, driver version etc. But the last time a user reported a problem like this was a LOOONG time ago. I had thought issues like this had been a thing of the past. In any case, there's probably not much I can do about it, except maybe allowing smaller 3D LUTs to be used.

i'm talking about 100 % more GPU usage.

but if the GPU is handling this i guess life suxxs.

edit:
here a screen showing the changing GPU usage :GPU usage (http://abload.de/img/gpuusageawsg9.png)

the GPU Vram issue comes from changing settings even with 90.15. for some reason it is not 100% freed up again.
so the main issue comes from the file with the unusual colors.

[leeperry]

using 115% fonts in W7SP1 I can't read the new antibloat thingie: http://thumbnails114.imagebam.com/47336/dc22e0473354680.jpg (http://www.imagebam.com/image/dc22e0473354680)

hopefully there isn't any further suboption underneath?

all those new suboptions are pretty subtle so it might take a while to test them all and some videos might benefit from a softer AR and others not, I would agree that SR doesn't ring to begin with but I'll try them all anyway :devil:

[baii]

As if it wasn't already confusing enough:
Even in FSE, I'm getting one dropped frame after 5 minutes of watching. So I disabled DXVA2 copyback in LAV to see if it makes any difference, and indeed: Now queues also get filled properly in windowed fullscreen and I can watch more than 5 minutes without any issues so far. However, still low level monitoring with driver must be used in Afterburner.
Hope they fix this at some point in the drivers with new GPUs (got the same problem with Sapphire Trixx OC tool instead of Afterburner) or else one might need to use an external GPU (e.g. of APU) for decoding if one wants to use voltage control or custom fan curve with HEVC 10 bit content.
I don't really have that problem with afterburner running, I did not touch the low level w.e setting though. Maybe try speedfan or older driver.
Personally I don't have any positive experience with any of the crimson drivers.



i'm talking about 100 % more GPU usage.

but if the GPU is handling this i guess life suxxs.

edit:
here a screen showing the changing GPU usage :GPU usage (http://abload.de/img/gpuusageawsg9.png)

the GPU Vram issue comes from changing settings even with 90.15. for some reason it is not 100% freed up again.
so the main issue comes from the file with the unusual colors.

I often run madvr on the edge of my GPU performance so i am curious in this as well. For 1080i60 I can have one scene spike to 17-20+ms render time when the whole clip avg take about 13-14ms.
This typically happen for me is TV broadcast, a 1080i60 blu ray would not have such problem. Does the bit rate matter? But that doesn't sound vright? Since it is still same number of pixels.
Sent from my 306SH

[AngelGraves13]

Bug report for 0.90.16

* chroma and image upscaling: activate anti-ringing filter is always set to "relaxed." Changing it has no effect as it always reverts to "relaxed" from "strict (soft)." Because of this, I am unable to test the benefits of the "strict (soft)" setting.

[vivan]

I've just upgraded from v0.89.x and now I'm getting occasional (~1 in 20-50 randomly opened files) crashes at startup (when opening new file in mpc-hc, but never happened when opening next file). It seems that this issue was introduced in v0.90.0 (0.89.19 doesn't crash while v0.90.0 does).
It happens randomly, and it doesn't seem to favor any specific files.

I don't get neither madVR or mpc-hc crash reporter, but maybe because I have visual studio installed (instead? windows asks me if I want to debug with it). Log is empty too (it does write it normally when it doesn't crash).
Switching to last stable mpc-hc, software decoding and restoring default settings didn't help.

I do have dumps, but those are saved using VS and are bigger than minidumps: 1 (https://dl.dropboxusercontent.com/u/16254258/d9/madvr/mpc-hc64.7z), 2 (https://dl.dropboxusercontent.com/u/16254258/d9/madvr/mpc-hc64_2.7z) (madVR v0.90.16 and mpc-hc v1.7.10 stable).

[madshi]

But madVR versions prior to 0.90.4 almost haven't had this issue...
That may be the case, but the changes I did had their purpose, and I don't like spending hours and hours trying to analyze something which only seems to affect one Windows XP user, who is 3 service packs behind. Please at least try if SP6 fixes the issue, before I dump any more of my time on this.

i'm talking about 100 % more GPU usage.

but if the GPU is handling this i guess life suxxs.

edit:
here a screen showing the changing GPU usage :GPU usage (http://abload.de/img/gpuusageawsg9.png)

the GPU Vram issue comes from changing settings even with 90.15. for some reason it is not 100% freed up again.
so the main issue comes from the file with the unusual colors.
So no 90.16 specific bug? If it's a general bug, that's useful to analyze, too, of course, but right now I'm mostly worried about new bugs introduced by 90.16.

using 115% fonts in W7SP1 I can't read the new antibloat thingie
Ok, not nice, but as mentioned in the release notes, many of these features are considered experimental and might go again already in the next build, depending on feedback. So I didn't feel like redesigning the whole settings page, just to make space for a new experimental option.

hopefully there isn't any further suboption underneath?
No.

all those new suboptions are pretty subtile so it might take a while to test them all and some videos might benefit from a softer AR and others not, I would agree that SR doesn't ring to begin with but I'll try them all anyway :devil:
Anti-bloating should be quite obvious, at least for sharpening and SSIM, when using strong sharpening and SSIM strengths. E.g. try increasing SSIM strength to 100% and then using the Anti-Bloating filter with 100% on top of that. You might like the result, and it's anything but subtitle in that configuration.

Bug report for 0.90.16

* chroma and image upscaling: activate anti-ringing filter is always set to "relaxed." Changing it has no effect as it always reverts to "relaxed" from "strict (soft)." Because of this, I am unable to test the benefits of the "strict (soft)" setting.
Already reported *and* replied to, just a few posts ago. This "relaxed" vs "strict" setting was never meant for chroma and image upscaling, so it doesn't matter that it doesn't work there. It's only meant for image downscaling. And there is works. So it's not a real bug, just a cosmetical issue.

[madshi]

I've just upgraded from v0.89.x and now I'm getting occasional (~1 in 20-50 randomly opened files) crashes at startup (when opening new file in mpc-hc, but never happened when opening next file). It seems that this issue was introduced in v0.90.0 (0.89.19 doesn't crash while v0.90.0 does).
It happens randomly, and it doesn't seem to favor any specific files.

I don't get neither madVR or mpc-hc crash reporter, but maybe because I have visual studio installed (instead? windows asks me if I want to debug with it). Log is empty too (it does write it normally when it doesn't crash).
Switching to last stable mpc-hc, software decoding and restoring default settings didn't help.

I do have dumps, but those are saved using VS and are bigger than minidumps: 1 (https://dl.dropboxusercontent.com/u/16254258/d9/madvr/mpc-hc64.7z), 2 (https://dl.dropboxusercontent.com/u/16254258/d9/madvr/mpc-hc64_2.7z) (madVR v0.90.16 and mpc-hc v1.7.10 stable).
Ouch. Can you say that this was *definitely* introduced in exactly 0.90.0 and didn't happen yet in 0.89.19? If so, I could build a couple intermediate builds between 0.89.19 and 0.90.0 to isolate the change which is responsible for the crash.

[Ver Greeneyes]

That may be the case, but the changes I did had their purpose, and I don't like spending hours and hours trying to analyze something which only seems to affect one Windows XP user, who is 3 service packs behind. Please at least try if SP6 fixes the issue, before I dump any more of my time on this.
Uh, madshi, Service Pack 3 was the last service pack for Windows XP (as can be seen in the MSDN downloads (https://msdn.microsoft.com/en-us/subscriptions/downloads/#searchTerm=Windows%20XP&ProductFamilyId=0&Languages=en&Architectures=x86&ProductFamilyIds=140&FileExtensions=.iso&PageSize=10&PageIndex=0&FileId=0)). There were of course many individual updates released afterward, but I don't know where you're getting 3 additional service packs from.

[CarlosCaco]

thanks madshi for the new features
its a lot of features to test, take some time to say something about...
i sugest to a future version a debug mode similar to the debug mode from avisynth unsharp mask HQ that turns the movie green and you can fine tune the sharpen seeing the dark areas (flat) and light areas and lines that are sharpened, you can cleary see ringin, noise etc...
its good mode to fine tune all features with the algorithms
think about that
thanks for the efforts

[madshi]

Uh, madshi, Service Pack 3 was the last service pack for Windows XP
Ooops. It was NT which had 6 SPs. I wasn't sure, so I had googled and found some hits for XP SP6, so I thought it existed. Sorry for the confusion... :o

[khanmein]

DXVA2-CB/native + HEVC + GTX 970 is not working well when playing h.265 .mkv during i tried to drag the seek-bar.

i tested NV CUVID is way more smoother.

[Dreamland]

anti-bloating is GREAT!!

great work!

[ryrynz]

madshi you said you were aware of the device profile problems with it only entering one thing at a time, will that be fixed up next version?

[omarank]

Here are the most important areas of FEEDBACK I need:

a) Do you consider the "anti-bloating" filter useful for sharpening (LumaSharpen + AdaptiveSharpen)? If so, which strength settings do you like?
Yes, at 75% strength setting. Actually I don’t use either of the two sharpeners, so I don’t have a strong preference here.

b) Do you consider the anti-bloating and anti-ringing options useful for SuperRes? I think anti-bloating might not be beneficial here, but I'd like to hear your opinion. Do you see any benefit in the anti-ringing option? SuperRes already didn't ring that much, so I wonder if the separate anti-ringing post processing option is worthwhile in your opinion or not?
I like the anti-ringing enabled and the anti-bloating disabled for SuperRes. I would prefer that the anti-ringing option stays.

c) Do you consider the "anti-bloating" filter useful for SSIM downscaling? If so, which strength settings do you like?
No. It tends to take away that “SSIM” look from the image.

d) Which of the two anti-ringing options do you prefer for downscaling? (Edit: FYI, these options do *not* affect SSIM downscaling)
Strict (soft). Now I am a bit confused between Jinc Strict AR LL and SSIM 1D 25% AR LL. The crisp image of the latter is very impressive but I also very much like the analog look of the former.

[Hyllian]

Hi, madshi, congrats for the new version. Is there any theory behind what anti-bloating does? I have some difficult to understand what's its purpose.

[chros]

It may vary with the scene because if the scene has "unusual" colors, that part of the 3D LUT might not be in the texture cache. A lot depends on how the GPU driver handles the 3D LUT, how much of it is in the texture cache etc. Will depend on GPU texture cache size, driver version etc. But the last time a user reported a problem like this was a LOOONG time ago. I had thought issues like this had been a thing of the past. In any case, there's probably not much I can do about it, except maybe allowing smaller 3D LUTs to be used.
Thanks for the info, I haven't known this.
i'm looking into that. i simply wanted to know if the change to super xBR is taking extra resources.

90.15 was working. even with a render queue of 6.

looks like it is 3d LUT related. i don't get rendering spike without it. but a 3D LUT should always take the same processing power with CFR content.

I often run madvr on the edge of my GPU performance so i am curious in this as well. For 1080i60 I can have one scene spike to 17-20+ms render time when the whole clip avg take about 13-14ms.
This typically happen for me is TV broadcast, a 1080i60 blu ray would not have such problem. Does the bit rate matter? But that doesn't sound vright? Since it is still same number of pixels.

I can also check this tomorrow, but for that: can you upload a 1-2 minute sample (that includes both good and bad part of the video) to some host to be able to replicate the problem? (You can cut it out with mkvtoolnix.)

[madshi]

DXVA2-CB/native + HEVC + GTX 970 is not working well when playing h.265 .mkv during i tried to drag the seek-bar.

i tested NV CUVID is way more smoother.
This sounds like a decoder specific topic, which doesn't belong into this thread.

anti-bloating is GREAT!!
Glad you like it! Are you talking about sharpening, though? Or about SSIM downscaling? Or SuperRes?

madshi you said you were aware of the device profile problems with it only entering one thing at a time, will that be fixed up next version?
I'm not 100% sure what you mean right now.

Yes, at 75% strength setting. Actually I don’t use either of the two sharpeners, so I don’t have a strong preference here.

I like the anti-ringing enabled and the anti-bloating disabled for SuperRes. I would prefer that the anti-ringing option stays.

No. It tends to take away that “SSIM” look from the image.

Strict (soft). Now I am a bit confused between Jinc Strict AR LL and SSIM 1D 25% AR LL. The crisp image of the latter is very impressive but I also very much like the analog look of the former.
Thanks!

FWIW, instead of SSIM 1D 25% without Anti-Bloating try combining SSIM 1D 100% with 50% or 75% Anti-Bloating. Try downscaling a sharp high-quality high-res foto to a *much* smaller resolution, e.g. 4x or 8x downscaling factor. It may change your mind.

Hi, madshi, congrats for the new version. Is there any theory behind what anti-bloating does? I have some difficult to understand what's its purpose.
Images say more than words. So here are a couple of comparison screenshots, showing what Anti-Bloating does when used with AdaptiveSharpen:

original (http://madshi.net/madVR/AS_AB/originalFrame.png) - | - no sharpening (http://madshi.net/madVR/AS_AB/withoutAnySharpening.png) - | - AdaptiveSharpen 1.5 (http://madshi.net/madVR/AS_AB/AS15.png) - | - AS 3.0 applied in 2x supersampling (http://madshi.net/madVR/AS_AB/SS_AS30.png) - | - AS 1.5 + Anti-Bloating 50% (http://madshi.net/madVR/AS_AB/AS15_AB50.png) - | - AS 1.5 + Anti-Bloating 100% (http://madshi.net/madVR/AS_AB/AS15_AB100.png) - | - AS 0.5 (http://madshi.net/madVR/AS_AB/AS5.png)

Basically if you compare the "no sharpening" with "AdaptiveSharpen 1.5" you should see that the image gets nicely sharpened, but gets a somewhat "bloated" look. Meaning everything looks simply a little bit fat / bloated. This is what you usually get by using unsharp masking. You can lessen this bloated look by using supersampling (upscaling + sharpening + downscaling), which you can see in the "AS 3.0 applied in 2x supersampling" image. Of course that costs a lot of extra GPU performance.

Now have a look at the "AS 1.5 + Anti-Bloating 50% / 100%" images and you should see that the 50% Anti-Bloating filter has a similar effect as using supersampling, at a much lower performance cost. And the 100% image now has nicely "thin" edges and features. The bloated look is totally gone. Now one might wonder whether Anti-Bloating simply reduces the amount of sharpening. But no, that's not the case, which you can double check by comparing the "AS 1.5 + Anti-Bloating 100%" image to the "AS 0.5" image.

[ryrynz]

I'm not 100% sure what you mean right now.


Another issue, when typing in a profile, group name & rules under devices each key press takes me to color and gamma, so I have to click back on the profile/group name etc after each key press to finish typing.


Yeah, it's on my to do list.




No. It tends to take away that “SSIM” look from the image.


I think I agree with this as well, need to look at some 4k downscaling rather than supersampled video to be sure.

[RDJ]

Subtitles now shift up instead of dropping off the screen with narrow windows when crop black bars is disabled and move subtitles is enabled.

When black bar detection is enabled but crop black bars is disabled, black bars still get cropped with short windows.

[Hyllian]

Images say more than words. So here are a couple of comparison screenshots, showing what Anti-Bloating does when used with AdaptiveSharpen:

original (http://madshi.net/madVR/AS_AB/originalFrame.png) - | - no sharpening (http://madshi.net/madVR/AS_AB/withoutAnySharpening.png) - | - AdaptiveSharpen 1.5 (http://madshi.net/madVR/AS_AB/AS15.png) - | - AS 3.0 applied in 2x supersampling (http://madshi.net/madVR/AS_AB/SS_AS30.png) - | - AS 1.5 + Anti-Bloating 50% (http://madshi.net/madVR/AS_AB/AS15_AB50.png) - | - AS 1.5 + Anti-Bloating 100% (http://madshi.net/madVR/AS_AB/AS15_AB100.png) - | - AS 0.5 (http://madshi.net/madVR/AS_AB/AS5.png)




Thanks for the explanation. The effect is very subtle, but noticeable.

Does it only works after sharpening filters? Could it be use after Super-xBR to prevent bloating? Did you test it?

[huhn]

I can also check this tomorrow, but for that: can you upload a 1-2 minute sample (that includes both good and bad part of the video) to some host to be able to replicate the problem? (You can cut it out with mkvtoolnix.)

i don't see how they are related.

and to reproduce my issue you need to scale to UHD and you need a 3d LUT. maybe even my 3d LUT.

at 1080p there are no spikes what so ever. so it could be to little Vram or something simple like that.

i can make a sample later but that's not that easy. because i get small spikes all the time.

[scollaco]

I'm a big fan of the anti bloating. It is great!! Currently for 1080p content, I'm using crispen edges 1.0, Lumasharpen 1.0 and both anti bloating and anti ringing checked.

Thanks Madshi!

[omarank]

FWIW, instead of SSIM 1D 25% without Anti-Bloating try combining SSIM 1D 100% with 50% or 75% Anti-Bloating. Try downscaling a sharp high-quality high-res foto to a *much* smaller resolution, e.g. 4x or 8x downscaling factor. It may change your mind.
I tried but I still have the same opinion as earlier. To my eyes, SSIM doesn’t have any bloating effect and so when anti-bloating is applied it’s thinning effect makes the images look kind of unnatural. Also SSIM has a characteristic look – crisp images with nice defined textures; that look goes away with anti-bloating. Having said that, I very much appreciate all the efforts that you put in providing new algorithms.


Images say more than words. So here are a couple of comparison screenshots, showing what Anti-Bloating does when used with AdaptiveSharpen:

original (http://madshi.net/madVR/AS_AB/originalFrame.png) - | - no sharpening (http://madshi.net/madVR/AS_AB/withoutAnySharpening.png) - | - AdaptiveSharpen 1.5 (http://madshi.net/madVR/AS_AB/AS15.png) - | - AS 3.0 applied in 2x supersampling (http://madshi.net/madVR/AS_AB/SS_AS30.png) - | - AS 1.5 + Anti-Bloating 50% (http://madshi.net/madVR/AS_AB/AS15_AB50.png) - | - AS 1.5 + Anti-Bloating 100% (http://madshi.net/madVR/AS_AB/AS15_AB100.png) - | - AS 0.5 (http://madshi.net/madVR/AS_AB/AS5.png)

Nice to see this comparison. Anti-bloating works very well with sharpening algorithms.

[chros]

i don't see how they are related.
Madshi said, that it can depend on the scene.
and to reproduce my issue you need to scale to UHD and you need a 3d LUT. maybe even my 3d LUT.

at 1080p there are no spikes what so ever. so it could be to little Vram or something simple like that.
Then I can't test it, I don't have an UHD display :)

[huhn]

Madshi said, that it can depend on the scene.
but there is no word that he is using a 3D LUT.
Then I can't test it, I don't have an UHD display :)

the more i try to reproduce this the more strange it gets.

XY SUBfilter is amplifying the issue. with it i can get 400% higher GPU usage most likely way more. i saw AVG rendertimes of over 120ms. with using only a 3D LUT it is only up 50-100% higher GPU usages.

even more strange thing is that this is happening even if no subs are displayed or if they are hide. so as long as xy subfilter is loaded with a 3D LUT it is amplifying the issue.

and most strange thing is that XY subfilter without a 3D LUT doesn't really affect performance.

[huhn]

here a sample with a 3D LUT.
http://www.file-upload.net/download-11420168/3DLUTslowdownissue.7z.html

after loading the 3D LUT you can toggle it with shift+control+3

i can't trigger this issue without using native UHD. zooming in at 1080p doesn't trigger it.

[chros]

here a sample with a 3D LUT.
http://www.file-upload.net/download-11420168/3DLUTslowdownissue.7z.html

after loading the 3D LUT you can toggle it with shift+control+3

i can't trigger this issue without using native UHD. zooming in at 1080p doesn't trigger it.
OK, I'll try it out tomorrow on my FullHD TV just in case.

[aufkrawall]

Images say more than words. So here are a couple of comparison screenshots, showing what Anti-Bloating does when used with AdaptiveSharpen:

original (http://madshi.net/madVR/AS_AB/originalFrame.png) - | - no sharpening (http://madshi.net/madVR/AS_AB/withoutAnySharpening.png) - | - AdaptiveSharpen 1.5 (http://madshi.net/madVR/AS_AB/AS15.png) - | - AS 3.0 applied in 2x supersampling (http://madshi.net/madVR/AS_AB/SS_AS30.png) - | - AS 1.5 + Anti-Bloating 50% (http://madshi.net/madVR/AS_AB/AS15_AB50.png) - | - AS 1.5 + Anti-Bloating 100% (http://madshi.net/madVR/AS_AB/AS15_AB100.png) - | - AS 0.5 (http://madshi.net/madVR/AS_AB/AS5.png)

Neat. That bloated look was always what kept me away from using other sharpeners than SuperRes.
However, with just SuperRes alone, Anti-Bloating seems not necessary to me.

[XMonarchY]

Anti-bloating seems to make downscaling with SSIM 2D 100% look like Catmull-ROM downscaling - softer, losing the very value of SSIM... It does look good when sharpening is applied first.

I'm more interested in whether SuperRes in Image Refinement looks better with new LL + AR settings enabled.