madVR - high quality video renderer (GPU assisted)

[ryrynz]

Originally Posted by pose

ryrynz, here is comparison from his own site. To me the after picture looks unnatural and with lots of details missing.
https://3.bp.blogspot.com/-wv05mfBis...rned%2Boff.png
https://1.bp.blogspot.com/-S6k4Cru_J...ening%2Bx4.png

Thanks. It's certainly further than I'd go, the edges aren't something you want to thin down considerably.. better to sharpen them and just thin them just a little.
The overall detail is still there there's actually little change to most of the image except for the lines, I don't understand the reason why people want to thin them so much, it ruin's the appearance of the characters.

Sometimes I found even thin edges at 1.0 went a bit overboard so I stopped using it and just rely on a Awarpsharpen, and a touch of line darkening through Avisynth along with some sharpening through madVR depending on the source.
I can make IMO 480 anime look similar to 720 content. It's all about having a firm yet gentle hand, it's easy to over do it.

[Georgel]

Originally Posted by ryrynz

Thanks. It's certainly further than I'd go, the edges aren't something you want to thin down considerably.. better to sharpen them and just thin them just a little.
The overall detail is still there there's actually little change to most of the image except for the lines, I don't understand the reason why people want to thin them so much, it ruin's the appearance of the characters.

Sometimes I found even thin edges at 1.0 went a bit overboard so I stopped using it and just rely on a Awarpsharpen, and a touch of line darkening through Avisynth along with some sharpening through madVR depending on the source.
I can make IMO 480 anime look similar to 720 content. It's all about having a firm yet gentle hand, it's easy to over do it.

Mostly because my screen in FHD and I want all anime to look FHD

Some algorithms are less needed in some anime that were natively produced in FHD, so I'm playing around with my settings a lot, but that image represents exactly why I love edge thinning.

It;s mostly about getting rid of blurry and too thick edges that make things look fuzzy. Not that much about thinning really, as much as about making edges look tangible, opposed to the gaussian blurred filtered look they have sometimes.

Well, I agree that sometimes it might be overkill and the image might degrade, so I think I might have to also step down a bit on the details.

I also think that for my images examples I'm using upscaling refinement -> edge thinning 8.0 while for image enhancements I'm stepping it down a lot, as it does break eyes of anime characters.

For real media content, I think that the sweet spot is (for me?) around 2.8 for edge thinning on image enhancements, whilst I'm still testing for upscaling refinement.

[Georgel]

Originally Posted by pose

ryrynz, here is comparison from his own site. To me the after picture looks unnatural and with lots of details missing.
https://3.bp.blogspot.com/-wv05mfBis...rned%2Boff.png
https://1.bp.blogspot.com/-S6k4Cru_J...ening%2Bx4.png

I don't think that details are missing. The backgrounds look less blurry really, even closer to a clear image

[Patrik G]

Originally Posted by huhn

what settings are you using and is your TV using interpolation at what refreshrate is it running?

can you amke a screen of the OSD and a screen with athe windows movie player?

settings doesnt matter as i have changed all and it doesnt affect the motion test.
i use 60Hz at 1080P
run the test for yourself and see what you come up with.
it runs at 60fps
http://privat.bahnhof.se/wb192876/FP...Resolution.mp4

but madvr and its smooth motion must be doing something good because motion on 24p content is way better (more fluid) than on the build in windows media player which this test runs smooth on.
so i think this motion test doesnt say it all about motion performance.
not for 24p content anyway.

but just for the papers

[huhn]

this test file is kind of broken.

it is interlanced and madVr isn't deinterlancing it because it is saying that is has 60 FPS before deinterlancing.

when lavfilter is deinterlancing than madVr things this file is 120 hz.

[a8213711]

I just discovered the results of Image Enhancements don't appear in the images saved with MPC-HC; are there many other functions which don't appear when saving images?

[huhn]

if i'm not mistaken nearly everything is not applied there.

[Asmodian]

Originally Posted by huhn

if i'm not mistaken nearly everything is not applied there.

Doesn't that function actually get the video before madVR has even done scaling? I think it grabs the frame after decoding but before madVR.

[huhn]

you still need to do a YCbCr->RGB conversation and madVR has to do that.

i have no clue if the selected chroma scaler is used for that or not.
if i remember correctly it is dithering even with disabled dithering(don't do this).

[Georgel]

Originally Posted by a8213711

I just discovered the results of Image Enhancements don't appear in the images saved with MPC-HC; are there many other functions which don't appear when saving images?

But the enhancements appear when you press print screen, if you want to notice differences, as far as I know.

[vivan]

Originally Posted by huhn

if i remember correctly it is dithering even with disabled dithering(don't do this).

AFAIK it dithers on CPU (using error diffusion).

[ibius]

Originally Posted by Asmodian

Doesn't that function actually get the video before madVR has even done scaling? I think it grabs the frame after decoding but before madVR.

'Crop black bars' gets applied to saved images, so madVR is active.

@madshi: speaking of cropping black bars, could we get file name tags for that, please?
Your last reply to leeperry on the subject didn't say whether you would implement it or not.

[ryrynz]

Originally Posted by Georgel

I don't think that details are missing. The backgrounds look less blurry really, even closer to a clear image

You quoted me for something I didn't say. I ran both through a picture comparison tool and there's little change in detail nothing really noticeable.. It's just the lines which we already know are thinned too aggressively.

Sent from my Nexus 5 using Tapatalk

[v0lt]

@madshi
I have a mk3d file with the following parameters:

MultiView_Count                          : 2
MultiView_Layout                         : Side by Side (left eye first)

Is there a way to transform this video to "row interleaved" format? With madVR of course.

[Warner306]

Originally Posted by v0lt

@madshi
I have a mk3d file with the following parameters:

MultiView_Count                          : 2
MultiView_Layout                         : Side by Side (left eye first)

Is there a way to transform this video to "row interleaved" format? With madVR of course.

madVR will only accept frame-packed 3D. So, no, it can't be converted.

[Stereodude]

Originally Posted by Warner306

madVR will only accept frame-packed 3D. So, no, it can't be converted.

SBS or TAB are just 2D, so of course madVR can accept them. It can output row interleaved too. However, I don't know if you can tell madVR that a 2D input video is a certain type of 3D and have it handle it accordingly.

[huhn]

that's what warner is saying.

in it's current state madVR only detects frame packed 3D as 3D and it can only convert these. more 3D features where planned AFAIK like the feature Volt is talking about reading some flags in mkv and convert the input to the selected output.

[Sunspark]

Originally Posted by ryrynz

I can make IMO 480 anime look similar to 720 content. It's all about having a firm yet gentle hand, it's easy to over do it.

The after image linked was too heavy a hand. If you compare the red and yellow bows they wear, especially the one in the center, the yellow portions have actually shrunk and are now smaller than they were before by several mm.

[fhoech]

Originally Posted by huhn

how does your 3d lut know the brightness of the file?

It doesn't. All that it knows is the HDR peak brightness (of the display), as set during 3D LUT creation, then applies the BT.2390 rolloff + saturation adjustment, then maps into the display colorspace (with imo great results btw, my implementation has come some way in the past three weeks and will be available in the next version of DisplayCAL). Of course, a 3D LUT cannot react to dynamic metadata, which could be seen as a drawback. So far, I don't think it matters much, though.

I'm afriad it's not in the current versions from v0.90.22:

Sure, I know that. I'm currently working around it by having two versions of the files: One with HDR metadata (handled by madVR's HDR implementation) and one without (HDR handled by the 3D LUT). The only thing that's a little bothersome when I do comparisons is that I have to remove the HDR 3D LUT when using madVR's HDR, otherwise the HDR processing is done twice (once from madVR itself and once through the 3D LUT).

You can try v0.90.20: if it's working there you can ask madshi to revert/modify that change.

When I downgrade I won't be able to easily do comparisons. Having a way to disable madVR's HDR processing but not disable the 3D LUT would be useful. Notifying @madshi.

[AngelGraves13]

For some reason I keep going back to Jinc for all my scaling needs. It seems to look more natural and has almost no artifacts unlike the other scalers.

NNEDI3 128 does look quite good. 256 looks a tiny bit better, but my 1080 GTX tends to get hot with that setting and the render times are at about 39 ms, which is too close to the frame time of 24p content. Should I lower the setting if the max sometimes hits 45 ms, but the average is around 30 ms? I'd ideally want the max under the 41.7 ms of 24p content?