madVR - high quality video renderer (GPU assisted)

[oudo86]

Originally Posted by mzso

You can't use 32 bit madVR with a 64 bit player, it's impossible.
MadVR includes both in one package. So If your player is using madVR it's using the appropriate version.

oh yes, of course. I don't know why I got so confused on that.

Thanks.

[omw2h]

I quite like the new updated AdaptiveSharpen, less ringing and less posterization as the old, grain not being as visible now and no fattening of dark lines.
Previous AS had less tolerance where the line is drawn witch is a good optiont to have in anime, did not crisper edges and the grain stayed where it belongs.

(as far as i have compared, i don't want to misrepresent)

Just different pros and cons for the end result... still wish for the previous one for anime though.

[JarrettH]

Are these new filters for poor sources or artifacts/blocking due to low bitrate? The development these last ~10 pages has been interesting, but I prefer just a touch of low debanding in place of potentially removing detail.

[ryrynz]

Not just poor quality sources, the vast majority of online/broadcast content out there benefit from these algorithms.

Originally Posted by Jtacdf

I cannot tell the difference from "don't reduce texture detail " on/off. If it does something better, I suggest to leave it on by default.

You need the right content.. I was looking at a 720 episode of The Orville and it was easily visible.

You're always going to lose detail when enabling these things, everyone has their own trade off limits. Don't think I'll ever use RCA over 2, but then I prefer high quality sources to start with.

[nevcairiel]

Originally Posted by ryrynz

Not just poor quality sources, the vast majority of online/broadcast content out there benefit from these algorithms.

If a source benefits from compression artifact removal, isn't it then by default poor quality?

[ryrynz]

I suspect his definition of poor quality was lower quality than the content that would stand to gain benefits, but yes.. Cheeky.

[ryrynz]

BTW you haven't weighed in on RCA yet Nev..

[x7007]

When I use Dithering - Error Diffusion
Should I need to select Use Colored noise ? because I don't remember if when choosing Ordered DIthering then I need to deselect them or select them , also the change dither for every frame, but this gives more noise, is that helpful ?

[ryrynz]

Trust your eyes, it's all personal preference. Most tend not to select colored noise I believe, changing dither every frame is useful if you can see the pattern.

[Anima123]

While applying RCA with strength 1 on a good 720p source, I found that 'don't reduce texture detail' off is doing a better job than on. It seems that this option might be good for cartoon materials, but not good for real-life videos.

[x7007]

Originally Posted by ryrynz

Trust your eyes, it's all personal preference. Most tend not to select colored noise I believe, changing dither every frame is useful if you can see the pattern.

I still can't see any differences between Error 1 and 2 , and also the other 2 options ... so I can't be sure. I don't know which scene I will see it clearly , if I had a movie or some part I could see the differences then it would be easy, but I don't even know what to look for.

[ryrynz]

Dithering expresses itself IMO as a kind of haze over the picture, certain panels and settings may expose some options to be more visible than others. This may help you choose.

[huhn]

jpg and not even the same frame.

this "guide"...

[x7007]

Originally Posted by ryrynz

Dithering expresses itself IMO as a kind of haze over the picture, certain panels and settings may expose some options to be more visible than others. This may help you choose.

Originally Posted by huhn

jpg and not even the same frame.
this "guide"...

What you guys choose for LG OLED55C6 ? I mean it's suppose to be one of the Error and my guess with the 2 options selected because the panel doesn't have any issues with those things, as my guess it shouldn't.

[kalston]

Originally Posted by huhn

jpg and not even the same frame.

this "guide"...

Yeah I like the idea behind this wiki but the picture comparisons are ridiculous.

Just make keyboard shortcuts for different profiles with different settings in madvr, pause playback and then switch between them. Or take lossless screenshots so you can easily zoom in/ put them side by side.

[Blackwalker]

Originally Posted by Blackwalker

I never thought it, sorry.
when home i'll do a sceen of madvr OSD and all you suggest,
thx huhn

hummm....hardware decoder in lavfilter? you mean in the mpc-hc , video decoder?
i'ìll do a screenshot too.


Later and thx again

here some screenshots:

mpc-hc settings:

https://imgur.com/a/M3cXU
https://imgur.com/a/e6koI

madvr settings:

https://imgur.com/a/dY2xT
https://imgur.com/a/FuGXa
https://imgur.com/a/TEHOg
https://imgur.com/a/HCybZ
https://imgur.com/a/sBbxF
https://imgur.com/a/pi2tb
https://imgur.com/a/VXJi6
https://imgur.com/a/GNavv
https://imgur.com/a/K7PdV
https://imgur.com/a/zAMHd
https://imgur.com/a/zAMHd
https://imgur.com/a/IAmK2
https://imgur.com/a/h6xv4


Cpu and ram usage with DXVA Copy-back on lav filters:

https://imgur.com/a/zf2rC

Cpu ,ram usage and info with DXVA native on lav filters:

https://imgur.com/a/eL6OE
https://imgur.com/a/MU3Cz


GPU usage and info with DXVA Copy-back on lav filters:
https://imgur.com/a/I9YeN

GPU usage and info with DXVA native on lav filters:

https://imgur.com/a/82E1V


Sorry but i try to put link on "image" options with no success!

[ryrynz]

Originally Posted by huhn

jpg and not even the same frame.

this "guide"...

Hey I said it might help, not that it was perfect :P

Originally Posted by x7007

What you guys choose for LG OLED55C6 ?

Earth to Asmodian..

[madshi]

General support:

Originally Posted by thewildsun

Thanks. I'm not using any 3rd party software like f.lux.

Here's the OSD screenshot: https://i.imgur.com/s2tMJWk.png

Edit: It might be a RAM-related issue. I noticed that RAM only ever increases while playing a video. That explains why restarting the video player fixes the problem temporarily. I'm not sure what to try next, though...

Edit2: Resetting madVR to default settings did seem to alleviate the issue (I was able to play a 20 minute video with no frame drops) but if the issue is RAM-related, then that may just be because the default settings are less RAM intensive.

Edit3: For chroma upscaling, I had been using Recon-soft with SuperRes 3. Turning off SuperRes seems to fix the issue -- I need to do more testing to see if this is a true fix or not.

RAM being an issue would explain why it only happens after a while of smooth playback. SuperRes doesn't sound to me as if it would explain higher RAM usage, though. Are we talking CPU RAM or GPU VRAM, btw?

Originally Posted by mparade

JRiver latest 32-bit version. I haven't tried any other player yet (I have paid for it, so there shouldn't be any bug).

It's an important question to figure out if it's a bug in madVR or in the media player, because depending on that different developers are responsible for fixing the bug. I can't fix bugs in JRiver. So please try a different player. I don't want you to switch from JRiver to another player, but trying a different player is the best way to test if the bug is in madVR or in JRiver.

Originally Posted by mparade

I have problems with "3D" profile rule as well.

The first "if" should already take care of 3D, so you don't have to do all those "and (not 3D)" in the other "if"s. But that is just superfluous, it shouldn't stop it from working. So it seems to me that the "3D" test doesn't actually work. Can anybody confirm that? madVR does show "3D" in the Ctrl+J OSD, right?

Originally Posted by Anima123

Madshi, what is the order of artifact removal algorithms, if over one single options have been used?

Well, as the order of them might have impacts on the final quality of the processed image sequences, will you plan to open the setting order selection for us, or you have a way to determine which order is optimal in most cases?

I don't think choosing the order is necessary/useful for you. First is reduce compression artifacts, then reduce random noise, then debanding, and last deringing.

Originally Posted by Oguignant

Hi, I have a couple of questions about creating a HDR 3LUT with Calman ... I apologize if it is not the appropriate forum!

- In Calman I must choose Colorspace: BT.2020 HDR & Gamma: ST 2084 HDR (PQ)?

- In madTPG there are a couple of options that I do not know to choose. What should I choose? My TV is LG OLED 2017, I think it has 1,000 nits.

I don't know what to choose in Calman, please ask in the Calman support forum about that. The madTPG options define which metadata is sent to the display. I don't actually know which data you should ideally use their. Again, this is more a question for calibration experts, which I am not.

Originally Posted by Oguignant

the latest release v0.92.7 works HDR with the last driver of Nvidia?

Reports suggest not, unfortunately. I'd choose an older Nvidia driver.

Originally Posted by Werewolfy

3) 1 doesn't do anything so it can be removed. But there's a BIG difference between strength 2 and 3. 2 does very little and 3 removes almost all the noise and a lot of details. Is it possible to have one or two steps between 2 and 3?

Yes, I suppose I can add a setting between 2 and 3.

Originally Posted by Blackwalker

i always used mpc-hc+madvr but playing 4k video "the film goes to shots", "is not smooth"...

According to your screenshots, the only way the decoder is fast enough is to use "DXVA native". Is playback not smooth that way, either? The madVR OSD (Ctrl+J) does seem to suggest that the queues are properly filled, so it should play smoothly?

Originally Posted by mzso

Nope. I checked. Hard to mix up since the filenames are different for 32bit.

I uploaded #12-13, but I can't say it looks different to before.

Actually, #12-13 now contain information for MPC-HC, which is useful! Still no information for LAV yet, though. Maybe you can make that work somehow, too?

Originally Posted by rivera

Double checked it - same result, 60Hz is selected in "Control Panel\All Control Panel Items\Display\Screen Resolution\Monitor", 50Hz is marked as "active" in madVR settings.

Originally Posted by rivera

Also, could you comment this issue too

My best guess right now is that madVR has your displays somehow confused. Probably madVR is working with a different display than the one you're actually using. Do you have a dual-monitor setup? If so, which mode are you using (mirror, clone, extend, etc)?

Originally Posted by clsid

Having the estimated bitrate available as a parameter would be a start. You are right, estimating the quality level would require a considerable amount of feedback. So perhaps better to revisit this idea once the dust has settled and the new algos have been tested more thoroughly.

Ok, will add that to my to do list.

Originally Posted by oudo86

I am currently using MadVR 32bit on Windows 10 64bit with Potplayer 64bit.

Should I be using the 64bit version of MadVR?

You can either use 32bit madVR+Potplayer, or 64bit madVR+Potplayer. Both is possible, but you can't mix. Software HEVC decoding is faster in 64bit. Otherwise there's not too much difference, I think.

[madshi]

AdaptiveSharpen:

Originally Posted by ABDO

1) unchecked,with linear light it reduce Sharpen effect.

Yes, it does, but you can counter by setting sharpness higher. But you prefer it unchecked, anyway?

Originally Posted by ABDO

2) New AdaptiveSharpen better than the old one, the old did ahuge sharpen edges with visible artifacts, so i never used it, but the new one do enhance detail very nice and sharpen it with no visible ringing in the same time.

K, thanks.

Originally Posted by thighhighs

I like new AdaptiveSharpen, i use it now
I tested LL vs NO-LL, all sources SD upscaled to 1080p with NGU-AA, AR, NO-AB. My preferences is LL unchecked for anime, because sharpness looks more accurate overal, dark lines looks more thin and sometimes the image has less sharpened artifacts (rare). But for movies i prefer LL checked, because edges looks more natural for me.
Anyway, differences with low strength, like default 0.5, not too big for me.

Argh. I really wanted you guys to pick either LL or not LL, because I want to get rid of the option.

Originally Posted by Neo-XP

Source / With LL / Without LL

Thanks. The main difference I see is the strength of sharpening, though. LL seems noticeably softer than GL. What happens if you adjust the sharpening strength to make the sharpened images look as near to each other as possible. Do you then still prefer LL over GL?

Originally Posted by ryrynz

It's much better! At low AS strengths it's hard to tell much of a difference between GL and LL, although as Neo-XP's screenshots have shown earlier GL becomes sharper the higher AS goes.
That ain't the only thing, lines are thinner too using GL. Will have to see if I prefer a lower setting with GL or a higher one with LL.

Yeah, that would be good to know.

Originally Posted by ryrynz

Slightly sharper in some areas and slightly softer in others, but visually looks rather similar to the 92.4 version, I'm happy with it, with this being the latest "corrected version" I'm for that recent code change bacondither added to AS, made a big difference even with low strengths.

FWIW, the new version is from igv, not from bacondither.

Originally Posted by Jtacdf

I like the latest adaptive sharpen with 0.92.7. Previous two iterations was not usable imo with anime content as it always produce ringing and aliasing on the line art. It's now very refined to the point where it can be used.

I prefer linear light for sharpening to be unchecked. It seems to produce slightly softer and noisier result.

K, thanks.

Originally Posted by omw2h

I quite like the new updated AdaptiveSharpen, less ringing and less posterization as the old, grain not being as visible now and no fattening of dark lines.
Previous AS had less tolerance where the line is drawn witch is a good optiont to have in anime, did not crisper edges and the grain stayed where it belongs.

(as far as i have compared, i don't want to misrepresent)

Just different pros and cons for the end result... still wish for the previous one for anime though.

Just to be sure there are no misunderstandings: "the old" on is the one from v0.92.5+6, and "the previous one" is the one from v0.92.4? How much better do you still like the v0.92.4 version than the v0.92.7 version?

Conclusions:

First of all thanks for the detailed feedback, it's quite helpful!

Generally, it seems people prefer "linear light" unchecked for Anime content, but checked for filmed material? I'm not too happy with this distinction because I was hoping we could all agree on removing the "linear light" option and choosing a fixed value for it.

Would it maybe make sense to let you test a "half linear light" option which is right in the middle between linear light checked and unchecked? Maybe that would satisfy everyone?

[madshi]

Reduce compression artifacts:

Originally Posted by ABDO

5) as @Neo-XP notice, With the "don't reduce texture details" option checked, there are some compression artifacts left in the image that are not processed.

Yes, I've seen the same thing here. However, your example is a really extreme case and I found that adding some RRN seemed to fix the issue with this specific image.

Originally Posted by ABDO

it seem (don't reduce texture detail ) option save some texture detail in normal life content, and it give good result to the whole picture.

Yes, it sometimes helps preserving some texture detail.

Originally Posted by Neo-XP

Yes, it already removes a lot of compression artifacts at strength 1 for me without the "don't reduce texture details" option checked. With the "don't reduce texture details" option checked, it does not do much.
At 2 it is too strong, it removes a little more compression artifacts, but smoothen the entire image, even with the "don't reduce texture details" option checked.

Yes, I double checked and the results are the same with the same image or others.

Ok, of course we can't let Mila Kunis be uglified by compression artifacts!

Originally Posted by Werewolfy

4) I find 1 pretty good actually. It removes a lot of compression artifacts without damaging the picture when "don't reduce texture detail" is unchecked. It's my new default setting for SD videos so please keep it.

5) I really don't like this new option. There's more compression artifacts and the details are not more well preserved.
Between strength 1 on and off, I prefer when it's off for the reasons mentioned above.
Between strength 1 without this option and 2 with it, I definitelty prefer 1 without it. The picture is less blurry.

K, thanks.

Originally Posted by leeperry

Yes please keep the current RCA@1 and you can enforce "don't reduce texture detail" if you like because who would want to do that on purpose anyway?

The "don't reduce texture detail" option might not be properly named. That's what the option *tries* to do, but it also results in some compression artifacts not being reduced. So it has advantages and disadvantages.

Originally Posted by ryrynz

Love the don't reduce texture detail option, this with RCA at around 1 or 2 is an absolute must for most low res content. WIN!

Originally Posted by ryrynz

Switch away IMO.

I don't understand this one. First you seem to love the "don't reduce texture detail" option, then you say "switch away IMO". That seems to contradict each other?

Originally Posted by ryrynz

Does anything think RCA 11 and 12 are actually worthwhile? It's like taking a fresh print and pouring water over it. Could those settings actually ever improve anything?

Try on these crazy images:

blockyMoon.png, blockyVma.png

Ok, so it's probably unlikely that whole videos are as blocky as that. So maybe RCA 11 + 12 will only be useful for VapourSynth/AviSynth guys to "fix" just a couple of frames.

Originally Posted by ryrynz

I like 1, please keep.

Ok.

Originally Posted by asmo42

My conclusion is that I generally prefer the old way i.e. the box unchecked. I don't see a huge difference in detail with it on or off. However it seems to get a lot less effective at removing artifacts. Honestly to me this option seem to almost work like a better less destructive version of RRN. So I find it useful and would be sad to see it go.

K, thanks.

Originally Posted by Anima123

While applying RCA with strength 1 on a good 720p source, I found that 'don't reduce texture detail' off is doing a better job than on. It seems that this option might be good for cartoon materials, but not good for real-life videos.

Conclusions:

Once again thanks for the helpful feedback!

My impression is that strength 1 should stay, but pretty much everyone wants to use strength 1 with the option "don't reduce texture detail" *un*checked. That's fine.

At higher strengths, it seems some people like the "don't reduce texture detail" option, and some don't. That's really bad, though.

I'm sorry to say, but the option "don't reduce texture detail" is *not* going to stay, because it's not just a simple flag somewhere in the algorithm. It's basically a completely different algorithm. Same with all the various "strength" settings! And I don't have the resources to keep both algorithms available, at all the different strength settings, and then add a new quality/speed selector, as well.

So we need to come to a decision whether to have "don't reduce texture detail" enabled or disabled. *However*, I'm willing to compromise by turning "don't reduce texture detail" on or off, depending on the selected strength. So e.g. we could have it off for strength 1-3, and have it on for strengths 4-12, or something like that.

What do you guys think?

(As indicated above, strengths 11-12 might not be useful for madVR, but it will be useful for VapourSynth/AviSynth in the future, so I want to keep it.)