madVR - high quality video renderer (GPU assisted)

[aufkrawall]

Originally Posted by AndreaMG

I play dvds under MadVR without the "qdvd.dll" trick simply by adding in MPCHC external filters Cyberlink video decoder and Microsoft dvd navigator.

Could you please explain this a little more complete?
I installed PowerDVD but couldn't find the DVD decoder in the DS filter list.

[TheLion]

Originally Posted by madshi

Well, I would much rather implement a *real* sharpening algorithm, to be honest. That's why I'm also reluctant offering so many more tweaks: It costs many hours of my time which I could instead spend on doing the "real thing". Personally, I don't like Darbee pop at all. I much prefer something like Didee's FineSharp. So if you ask me to add this and that, and a couple of other tweaks in between, basically you're keeping me busy with so-so stuff, instead of allowing me to move on to bigger and greater things. I've said it so many times before: Let me first add all the missing features, before spending my time by twiddling with 1% improvements. I'm feeling held back at the moment, by all the constant requests for small tweaks and changes everywhere. Aren't you guys interested in getting things like custom shader support, 3D support etc?

I am looking forward to features like a sophisticated "real" sharpening filter, custom shader support and 3D support so badly that I will refrain from making any more requests/proposals I am very confident that your choice of development priorities is best for all of us.

[Blight]

Madshi:
I would also welcome 3D support in the renderer, if there's anything I can do to help push it forward, let me know.

[AndreaMG]

@aufkrawall, in MPCHC disable all internal filters (both source and transform), enable audio switcher and in "external filters" add:

- LAV Splitter (it will not be triggered by dvds...)
- LAV Video decoder (same as previous, currently not supported, but NEV is working at dvd support now)
- Cyberlink video/SP Video decoder (mine is PDVD10, as far as I know it is sufficient you install a trial version then the decorer remains there fo you
- ffdshaw raw video filter (required by SVP, which for me is a MUST)
- LAV Audio decoder (supported by dvds)
- Microsoft dvd navigator

This set up allows me to play any video but dvds using LAV Video (because it is set at top in the merit list), and only for dvds automatically MPCHC switches to Cyberlink Video decoder: no more problems with freezings when seeking and displaying menus.

Hope it helps

[NicolasRobidoux]

Originally Posted by madshi

FWIW, when comparing Jinc 3 vs 4 taps I can't see much of a difference when using my anti-ringing filter. 4 taps is ever so slightly sharper, that's all the difference I can see. I can't really see a different character.

I just checked, and I would guess that the anti-ringing reduces the differences. When I compared the two long long ago in a distant galaxy, the EWA Lanczos 4 halo made me puke. It's not only that the second halo is really obvious. The first halo is also deeper, esp. where straight, high contrast lines, meet.
Some examples are found here: http://web.cs.laurentian.ca/nrobidou.../madVR/glyphs/
They show EWA LanczosSharp (3-lobes) and results obtained with 4 and 5 lobes. To keep things simple, I used the same deblur for all three enlargements, which each correspond to a different number of lobes (the same one as madVR, and the same one as the default EWA LanczosSharp in ImageMagick), and I enlarged sRGB pixel values directly (no colorspace shenanigans).
P.S. There may be something broken with my ImageMagick install. I may have to update the images later.
P.S.2 No: It's Firefox making things look strange.
P.S.3 Now I remember what made me go "these schemes look really different" (besides the haloing): I was using very strong deblurs, roughly 0.885, in the hope of gettting something as sharp as tensor Lanczos. When deblurring this much, the three schemes look really different. (Again, this is from memory.) So, my comments about them having a different character may have to do with pushing the EWA Lanczoses where, really, they should not go.
-----
The parking meter image is fairly soft. I vaguely remember someone mentioning that it was "obvious" that it had been obtained by downsampling with box filtering, and I remember looking carefully and agreeing.
When enlarging blurred images, decent windowed Sinc and Jinc methods will generally give roughly similar results.
Are video images typically that soft and similar to the results of downsampling by box filtering?

[edragon]

Originally Posted by AndreaMG

Try without SVP, which eats a lot of CPU, but also GPU if enabled to do so. I have a 460GTXM overclocked and with both chroma and luma upscale Jinc 3 taps and SVP I have drop frames...

But SVP is very important for me , i can not watch anime without it .... though it really eats a lot of CPU and GPU!
So now i just use jinc3 without antiR ... but still have good image when upscalng 480p to 1080p

[aufkrawall]

Originally Posted by AndreaMG

@aufkrawall, in MPCHC disable all internal filters (both source and transform), enable audio switcher and in "external filters" add:

- LAV Splitter (it will not be triggered by dvds...)
- LAV Video decoder (same as previous, currently not supported, but NEV is working at dvd support now)
- Cyberlink video/SP Video decoder (mine is PDVD10, as far as I know it is sufficient you install a trial version then the decorer remains there fo you
- ffdshaw raw video filter (required by SVP, which for me is a MUST)
- LAV Audio decoder (supported by dvds)
- Microsoft dvd navigator

This set up allows me to play any video but dvds using LAV Video (because it is set at top in the merit list), and only for dvds automatically MPCHC switches to Cyberlink Video decoder: no more problems with freezings when seeking and displaying menus.

Hope it helps

Thanks, but unfortunately it doesn't work for me.
With PDVD 10, now the filters appear in the list (with 12 they don't) but I still get Macrovision error.

[NicolasRobidoux]

Originally Posted by madshi

I'm looking forward to that. However, I'm now using a very small Jinc LUT for my pixel shaders, using linear LUT interpolation, and it works surprisingly well. For your interest: I started with a LUT with 8 * 1024 values. Then I lowered the LUT size and compared images. I was surprised how small the LUT could be without getting any visible differences. I went down as far as 32 values only and still couldn't see any difference with bare eyes, although in a hex editor I could see some colors changed by a value of 1 (in 8bit). I'm now using a LUT size of 64 values. GPUs seem to handle that really well (due to texture caching), so it will be interesting to see whether your low flop math will be any faster.

Suggestion: Base the LUT on EWA_disc_radius^2 - distance^2. This saves square roots, and makes the LUT denser where it should (near the center) at no cost.
P.S. My guess is that with this trick, and offsetting sampling locations just a bit so that the "outermost" one does not have a weight equal to 0, you may be able to get away with 32 with no worries whatsoever? 16 even?
Computing the fast approximations just went down the priority list.

[Budtz]

Originally Posted by AndreaMG

And most of all it would be awesome to able to sharpen AFTER resizing, which is impossible via avisynth in conjunction with MadVR. In my opinion the integration (with pixel shaders?) of a sharpener in MadVR is priority N. 1

for me as well. i second that

[madshi]

Originally Posted by Blight

Madshi:
I would also welcome 3D support in the renderer, if there's anything I can do to help push it forward, let me know.

It's on my to do list, but there are some other things with higher priority. But it isn't too far away, I hope...

Originally Posted by NicolasRobidoux

I just checked, and I would guess that the anti-ringing reduces the differences. When I compared the two long long ago in a distant galaxy, the EWA Lanczos 4 halo made me puke. It's not only that the second halo is really obvious. The first halo is also deeper, esp. where straight, high contrast lines, meet.
Some examples are found here: http://web.cs.laurentian.ca/nrobidou.../madVR/glyphs/
They show EWA LanczosSharp (3-lobes) and results obtained with 4 and 5 lobes. To keep things simple, I used the same deblur for all three enlargements, which each correspond to a different number of lobes (the same one as madVR, and the same one as the default EWA LanczosSharp in ImageMagick), and I enlarged sRGB pixel values directly (no colorspace shenanigans).
P.S. There may be something broken with my ImageMagick install. I may have to update the images later.
P.S.2 No: It's Firefox making things look strange.
P.S.3 Now I remember what made me go "these schemes look really different" (besides the haloing): I was using very strong deblurs, roughly 0.885, in the hope of gettting something as sharp as tensor Lanczos. When deblurring this much, the three schemes look really different. (Again, this is from memory.) So, my comments about them having a different character may have to do with pushing the EWA Lanczoses where, really, they should not go.

Interesting. With that image, when using Jinc upscaling, my anti-ringing algorithm removes most of the halos, but it has ugly harsh edges. I'll have to see if I can fix that. FWIW, the anti-ringing filter doesn't look harsh at all with Lanczos. Wonder why that is, will have to check...

FWIW, with that image, with anti-ringing enabled, Jinc 3-taps and 4-taps look very similar.

Originally Posted by NicolasRobidoux

The parking meter image is fairly soft. I vaguely remember someone mentioning that it was "obvious" that it had been obtained by downsampling with box filtering, and I remember looking carefully and agreeing.
When enlarging blurred images, decent windowed Sinc and Jinc methods will generally give roughly similar results.
Are video images typically that soft and similar to the results of downsampling by box filtering?

Different film studios are using different downscaling methods. Some use things like Lanczos. Others appear to use Nearest Neighbor sampling (shudder). I guess some are using box filtering. Many DVDs are *very* soft, much softer than the parking meter image, and still sometimes have ringing artifacts in them, despite being so soft. Good Blu-Rays may be a bit sharper than the parking meter image, but I doubt they're much sharper. Generally I think movies are softer than photos.

I don't remember which method I used exactly to create that parking meter image, it was a couple years ago. It is a downscale, but I don't remember which algorithm I used. I think I used one which didn't ring (at all) because I wanted to have a source image with not even a hint of ringing, so I could properly evaluate which upscaling algorithms add ringing and which don't.

[AndreaMG]

Originally Posted by aufkrawall

Thanks, but unfortunately it doesn't work for me.

Strange, have you AnydvdHD enabled? It should remove macrovision...

[digitech]

Originally Posted by sunnah

Chroma upscaling: Jinc 8 taps
Image upscaling: Jinc 8 taps
Image downscaling: Lanczos 8 taps

(I have the anti-ringing option enabled on everyone of them.

All work great on the NVIDIA GTX 570M without dropped/delayed frames. The image and colours of 1080p, 720p and below are magnificent. And the display modes option also works really well for me.

Keep up the good work Madshi.

Sunnah im interested about your HTPC, can you give me details about processor type, speeed, amount of ram etc, etc, pls? im about to build a new htpc just for the main use of playing movies with madvr, any info about your htpc details will be very apprecciated, thanks!

[NicolasRobidoux]

Originally Posted by madshi

... Generally I think movies are softer than photos. ...

This is very important information.

[NicolasRobidoux]

I've added (tensor) Ginseng, Lanczos, Lanczos4, and EWA QuadraticJinc, produced similarly to the earlier ones, to http://web.cs.laurentian.ca/nrobidou.../madVR/glyphs/
-----
Mathias:
Ginseng is very often within JND of Lanczos, so much so that you probably could substitute it for Lanczos, and nobody would know better. (Sorry: I should have mentioned this earlier.) This being said, it does have measurably less halo, and it is slightly less jaggy, although the difference is only noticeable with sharp high contrast features. Although the difference is small, it is (on my brash Samsung monitor), noticeable on the "glyph" enlargements.
Tensor Lanczos is definitely too sharpening when enlarging. Ginseng backs the dial down from 11, just a nudge.

[NicolasRobidoux]

Mathias:
Indeed, the one advantage of QuadraticJinc (which I am oh! so tempted to christen the "Kaching" filter even though it is oh! so tacky) over LanczosSharp is that it has much reduced halo, without this costing too much in terms of jaggies, sharpness etc (as it does with Keys cubics, which are the best cubics).
So, if you manage to fix the anti-ringing filter for Jinc (or you go with sigmoidization, which does work better with the EWA Lanczoses than the tensor ones anyway), I would guess there is no need for it. Otherwise, use it for chroma when using Jinc for luma?

[madshi]

Thanks, Nicolas.

Do you have any pointers where I can find information about how the "full" sigmoidization works? I suppose it's higher quality than the fast "trick" you mentioned, right? If I start doing funny stuff with color space to improve scaling, I can just as well go the full mile and implement full sigmoidization (if it's not too complicated/slow).

[Xaurus]

Originally Posted by madshi

Well, I would much rather implement a *real* sharpening algorithm, to be honest. That's why I'm also reluctant offering so many more tweaks: It costs many hours of my time which I could instead spend on doing the "real thing". Personally, I don't like Darbee pop at all. I much prefer something like Didee's FineSharp. So if you ask me to add this and that, and a couple of other tweaks in between, basically you're keeping me busy with so-so stuff, instead of allowing me to move on to bigger and greater things. I've said it so many times before: Let me first add all the missing features, before spending my time by twiddling with 1% improvements. I'm feeling held back at the moment, by all the constant requests for small tweaks and changes everywhere. Aren't you guys interested in getting things like custom shader support, 3D support etc?

I fully agree with you, now that you added Jinc I think the algorithm-people should be satisfied.

For me I think 3D-support would be my nr. 1 on the wishlist.

[Xaurus]

Originally Posted by Moony349

For LAV I check off all of the output formats and leave RBG output levels as untouched.

To check your colorspace, what I've found works best is downloading the mp4 version of these test videos:

http://www.avsforum.com/t/948496/avs-hd-709-blu-ray-mp4-calibration

Use the APL clipping (Basic Settings #2) to make sure your colors are correct. Nvidia cards notoriously will output 16-235 over HDMI unless you hack the monitor INF or use Madshi's madnvlevelstweaker (I haven't tested it but I assume it works).

You want to be able to see blinking in the white area up to at least 234 (And black down to 17 or so ideally). If the whole screen is blinking your card is probably converting down to 16-235 and you should try using madshi's tweak.

I have to try that. I have only checked NV12 since I am certain nevcairiel (or was it madshi?) said that was the smartest thing to do.

I am already using the inf-hack for a time, and it works. The APL-test behaves like you describe.

[NicolasRobidoux]

Mathias:
I'm the one who invented sigmoidization, and when I program stuff into ImageMagick, there are certain things I can't do because I have to fit whatever I do into the existing structure.
-----
Can you point me to an original of your parkmeter image and what size enlargements you like to look at?
-----
You can see examples of sigmoidization at the following various places. Everywhere, it's used on all three channels, but I think that it may be better to only use it on luma or luminance.
If you try it yourself with ImageMagick, please use a very recent version, ideally of ImageMagick 7, because there was a bug in the IM sigmoidal-contrast code that was introduced, I think, in February, that I had to fix before I could use something else than the "sRGB pretend hack". I fixed it in ImageMagick 7 in August, and in ImageMagick 6 in September.
Look at this stuff with the suspicious eyes of someone who is presented the most beautiful baby in the world by his father.
http://www.imagemagick.org/Usage/res...las_upsampling
http://www.imagemagick.org/Usage/res...size_sigmoidal
http://www.imagemagick.org/discourse...p?f=22&t=21804
http://www.imagemagick.org/discourse...hp?f=1&t=21695
http://www.imagemagick.org/discourse...p?f=22&t=21933
http://www.imagemagick.org/discourse...hp?f=1&t=21850
http://www.imagemagick.org/discourse...p?f=22&t=21415

[nevcairiel]

Originally Posted by Xaurus

I have to try that. I have only checked NV12 since I am certain nevcairiel (or was it madshi?) said that was the smartest thing to do.

With madVR, the smartest thing to do is keep everything checked. Defaults, use them.