madVR - high quality video renderer (GPU assisted)

[huhn]

I've been messing with the settings trying to get the best setup for 720p->1080p NNEDI3 doubling, and I was noticing something was off but I couldn't pinpoint it until now. No matter what algorithm I use for downscaling, I get aliasing with text if anti-ringing is used. Anything else and text looks great, even Lanczos4 (no AR). Has anyone else noticed this or knows why this happens? I don't see any more ringing with it turned off, so I'm just going to leave it off for now.

can you provide a short sample where this is happing?

Asmodian: I'm not talking about playing native 59 fps at 29 or 59, 59 would be optimal and that's what all those samples (except mousse_smooth_big.avi is kind of close, duplicate frames to go from 24 to 50) on 100fps is discussing. I'm talking about native 29 fps played at 59 fps by duplicating every frame. Is there some advantage to it or is it just a waste of gpu and would be better off just playing at 29 fps?

if DEINT is used on a 2:2 or 3:2 cadence clip it BOB/line double each field and the inverse telecine/pulldown detection from nvidi/amd isn't working then you lose picture informations.

if MadVR IVTC is used you can easily watch them as 29/25/23 or what ever it is after IVTC no harm done

Here's a native 29 fps video. Screenshots of a 29 fps frame is bit identical to the 2 59 fps frames.

this should happen with IVTC with DEINT this is the case http://forum.doom9.org/showthread.php?t=170420

[Asmodian]

Originally Posted by turbojet

Asmodian: I'm not talking about playing native 59 fps at 29 or 59, 59 would be optimal and that's what all those samples (except mousse_smooth_big.avi is kind of close, duplicate frames to go from 24 to 50) on 100fps is discussing. I'm talking about native 29 fps played at 59 fps by duplicating every frame. Is there some advantage to it or is it just a waste of gpu and would be better off just playing at 29 fps?

Nevermind the video sample above, it played in wtv as 29 and 59 fps but detects as 3:2 in mpg. I think it's a potplayer issue, resolution is detected wrong too.

Here's a native 29 fps video. Screenshots of a 29 fps frame is bit identical to the 2 59 fps frames.

This could be about a 50% performance gain with the relatively common native 29 fps videos when double rate deinterlacing is enabled. Another large performance gain could be checking 50 and 59 fps for film. It's quite common to see film in 720p59. Just giving madshi some ideas if he hasn't already thought about it. i can provide samples when the time comes.

100fps.com is talking about normal 25i video, it has 50 points in time at half vertical resolution... that is the definition of 25i. It is the same for 29i. All those examples are after deinterlacing, they started with 25i. Except mousse_smooth_big.avi should have been IVTC'ed back to 24 fps instead of deinterlaced.

Your sample seems to be a native 59 fps progressive? I do not see any dup frames. Are you talking about real interlaced material or telecined material? In my experience native 29 fps progressive is pretty rare.

Talking about real interlaced 29 fps video; you need to take it to 59 to deinterlace without losing information. Half rate is an option if you want to use it.

A film in 720p59? I assume it was telecined and then double rate deinterlaced? Nasty and sounds hard to IVTC, the deinterlacing would mess that up.

[DragonQ]

Originally Posted by turbojet

Asmodian: I'm not talking about playing native 59 fps at 29 or 59, 59 would be optimal and that's what all those samples (except mousse_smooth_big.avi is kind of close, duplicate frames to go from 24 to 50) on 100fps is discussing. I'm talking about native 29 fps played at 59 fps by duplicating every frame. Is there some advantage to it or is it just a waste of gpu and would be better off just playing at 29 fps?

Nevermind the video sample above, it played in wtv as 29 and 59 fps but detects as 3:2 in mpg. I think it's a potplayer issue, resolution is detected wrong too.

Here's a native 29 fps video. Screenshots of a 29 fps frame is bit identical to the 2 59 fps frames.

This could be about a 50% performance gain with the relatively common native 29 fps videos when double rate deinterlacing is enabled. Another large performance gain could be checking 50 and 59 fps for film. It's quite common to see film in 720p59. Just giving madshi some ideas if he hasn't already thought about it. i can provide samples when the time comes.

If the frames are identical then the GPU's cadence detection is working and everything's fine. I can think of two disadvantages to doing this compared to using MadVR's IVTC: it presumably increases GPU usage since the video is now 50 fps instead of 25 fps, and it might affect the result of Smooth Motion. The main advantage I can think of is that you don't have to manually change settings when switching between interlaced and PsF video.

Originally Posted by Asmodian

100fps.com is talking about normal 25i video, it has 50 points in time at half vertical resolution... that is the definition of 25i. It is the same for 29i. All those examples are after deinterlacing, they started with 25i. Except mousse_smooth_big.avi should have been IVTC'ed back to 24 fps instead of deinterlaced.

Your sample seems to be a native 59 fps progressive? I do not see any dup frames. Are you talking about real interlaced material or telecined material? In my experience native 29 fps progressive is pretty rare.

Talking about real interlaced 29 fps video; you need to take it to 59 to deinterlace without losing information. Half rate is an option if you want to use it.

A film in 720p59? I assume it was telecined and then double rate deinterlaced? Nasty and sounds hard to IVTC, the deinterlacing would mess that up.

The video is encoded as 720p/60 but if you frame step it's obvious that it's actually from a 720p/30 source. Nothing unusual about that if it's from a broadcast though, happens all the time.

[huhn]

Originally Posted by DragonQ

The video is encoded as 720p/60 but if you frame step it's obvious that it's actually from a 720p/30 source. Nothing unusual about that if it's from a broadcast though, happens all the time.

the sample is true 59 fps the dup frame you see ware from the deint 59 = 120 look at the name it force deint on it.
the clip is totally right deinterlaced and should be treated as progressive.

[Asmodian]

Originally Posted by DragonQ

The video is encoded as 720p/60 but if you frame step it's obvious that it's actually from a 720p/30 source. Nothing unusual about that if it's from a broadcast though, happens all the time.

Originally Posted by huhn

the sample is true 59 fps the dup frame you see ware from the deint 59 = 120 look at the name it force deint on it.
the clip is totally right deinterlaced and should be treated as progressive.

Oh yeah, I did rename it. Deinterlacing was off and I stepped through single frames. The pan over the ceiling fan near the end was fairly convincing; no dup progressive 59.94 fps. I too suspect it is well deinterlaced 720i29.976.

[DragonQ]

Originally Posted by huhn

the sample is true 59 fps the dup frame you see ware from the deint 59 = 120 look at the name it force deint on it.
the clip is totally right deinterlaced and should be treated as progressive.

Ah yes I forgot about file name tags.

[DragonQ]

Originally Posted by Asmodian

Oh yeah, I did rename it. Deinterlacing was off and I stepped through single frames. The pan over the ceiling fan near the end was fairly convincing; no dup progressive 59.94 fps. I too suspect it is well deinterlaced 720i29.976.

720i isn't any kind of standard. If it's been deinterlaced it's been done from 1080i/30 almost certainly.

[Asmodian]

Originally Posted by DragonQ

720i isn't any kind of standard. If it's been deinterlaced it's been done from 1080i/30 almost certainly.

Ah, true. I was only thinking about the frame type and rate.

[tp4tissue]

Originally Posted by nevcairiel

He uses nearest neighbor and calls it perfect scaling, he is beyond help.

Aliased?

That's irrelevant when you get 1:1 scaling.. from 1280x720 to 2560x1600(1440)

That is as much information as is in the file.. anything ONTOP of that is an approximation.

I wouldn't use NN for 19x10 to 25x16... but that's only because NN shimmers.

[huhn]

Originally Posted by dansrfe

For some reason certain files exhibit a choppy playback but don't show dropped frames or decreasing queues. I'm thinking it's not a bug but should I post a report?

maybe a sample first?

[MokrySedeS]

Originally Posted by tp4tissue

Aliased?

That's irrelevant when you get 1:1 scaling.. from 1280x720 to 2560x1600(1440)

That is as much information as is in the file.. anything ONTOP of that is an approximation.

I wouldn't use NN for 19x10 to 25x16... but that's only because NN shimmers.

Here, a quick comparison: http://screenshotcomparison.com/comparison/70195
That's @ 200% zoom of course.

[6233638]

I think tp4tissue's point is that 720p videos scaled to exactly 2x size on a 2560x1440 or 2560x1600 display using Nearest Neighbor, should look exactly the same as if you displayed that 720p video 1:1 mapped on a 720p native display.

Which - aside from the change in pixel density - is true.
(the increased pixel density means you will actually have a sharper image, and are able to resolve finer details)

But unless you are displaying content which more-or-less requires pixel-perfect accuracy to look good (video games or a lot of PC software) I don't know why you wouldn't take advantage of the display's additional resolution and upscale via madVR.

[Fullmetal Encoder]

Originally Posted by 6233638

I think tp4tissue's point is that 720p videos scaled to exactly 2x size on a 2560x1440 or 2560x1600 display using Nearest Neighbor, should look exactly the same as if you displayed that 720p video 1:1 mapped on a 720p native display.

Which - aside from the change in pixel density - is true.
(the increased pixel density means you will actually have a sharper image, and are able to resolve finer details)

But unless you are displaying content which more-or-less requires pixel-perfect accuracy to look good (video games or a lot of PC software) I don't know why you wouldn't take advantage of the display's additional resolution and upscale via madVR.

Yes but aren't you getting quite a bit of blurring/degradation of the image in scaling up to 2560, even from 1920x1080? I don't prefer NN myself but I've been avoiding going up to 2560x1440 so that I can maintain 1:1 mapping while avoiding that at decent screen sizes. Although I've never actually had an opportunity to play with it on a 2560 screen.

With 4k screens coming I shudder to think what scaling up DVD content would look like. At this rate everyone will need two different screens to enjoy all of their content properly.

[turbojet]

I don't know how you guys are seeing 59 fps from that 29i source. It duplicates every frame which is unnecessary gpu load for madvr but that's how it currently works, would be nice for madvr to handle it better at some point.

Another case where madvr uses unnecessary gpu load is film in 59p.

Here's 2 cases, and the only cases where madvr should be outputting 59 fps: 59p and 59i.

[ryrynz]

I'd really like to see some more development put into NNEDI to attempt to improve it, I wonder the chances of it improving in future? Basically it sounds like someone needs to create an external Darby-like neural network upscaler.

I assume it wouldn't be cost effective to implement a good upscaler into normal devices, it's possible good quality lower res content upscaling won't get improved treatment with 4K devices at all.

I assume MadVR will be the best solution of a long time to come.

[nevcairiel]

Originally Posted by turbojet

I don't know how you guys are seeing 59 fps from that 29i source. It duplicates every frame which is unnecessary gpu load for madvr but that's how it currently works, would be nice for madvr to handle it better at some point.

I can only repeat what various people have told you before:

29i is supposed to be output as 59p. (Note that many people mix 29i and 59i, but generally mean the same thing, 29 interlaced frames/59 fields per second, ie. what is typically broadcast in NTSC countries)
If its interlaced, it should be output with twice the "frame rate". Its as simple as that.

That sample however is NOT 29i, its 29p. You renamed it with a "deint=video" tag in the filename, which FORCES madVR to deinterlace the video. If you do that, its going to treat it like a normal interlaced file, which as a result does indeed duplicate all frames - but thats only because you forced it to do that, it wouldn't do that on its own.

Interlaced video does NOT contain frames, it contains fields. For technical reasons, two fields are packed into one "frame". So what you do when deinterlacing is first split these two fields again - so your 29 fps interlaced stream becomes 59 fields per second, and then every field is individually deinterlaced, which results in 59 frames per second. Thats how proper deinterlacing HAS to work. If a stream is interlaced, that is what will ALWAYS happen.

[turbojet]

You are right nevcairel I need to stop relying on playing within a wtv container, madvr deinterlaced that same source and duplicated frames, it also showed a resolution of 720x480 even though it's really 29p inside of 720p59.

Here's the type of source I'm talking about, madvr finds cadence and displays mostly correct now. I remember months, maybe a year or so now it wasn't and I had to enable half frame deinterlacing to display things like this correctly. Which is an improvement and seems only film in 59p is the only unnecessary performance hog.

[Asmodian]

Originally Posted by nevcairiel

That sample however is NOT 29i, its 29p. You renamed it with a "deint=video" tag in the filename, which FORCES madVR to deinterlace the video. If you do that, its going to treat it like a normal interlaced file, which as a result does indeed duplicate all frames - but thats only because you forced it to do that, it wouldn't do that on its own.

Actually that sample is 59p and the forced deint takes it to 120 with duplicates. Same effect though.

[tp4tissue]

Originally Posted by MokrySedeS

Here, a quick comparison: http://screenshotcomparison.com/comparison/70195
That's @ 200% zoom of course.

720p to 25x16 or 15x14 using NN is akin to LOSSLESS conversion

720p to 25x16 or 15x14 + jinc/lanz/xxxx is Lossless + obvious-POSTPROCESSING... you guys remember those Emulator filters, same concept.


Granted madvr does lots of stuff behind the scene even with NN that would constitute post-processing.

But I'm am merely conveying my preference for prioritizing True-To-Source.


Not saying everyone should follow suit, or that it's the-right-way..

I honestly don't know why few o'yall acting like I just ate your lunches from the community freezer.



as with 19x10 to 25x16, well, I don't have a choice, my monitor is 25x16, I have to use the blur filters, or else everything looks blotchy (blotchy, not blocky)...

I like blocky if it's 1:1

[Asmodian]

Originally Posted by tp4tissue

Granted madvr does lots of stuff behind the scene even with NN that would constitute post-processing.

Any examples? If you have debanding off and are using NN, what else is there?

Originally Posted by tp4tissue

720p to 25x16 or 15x14 + jinc/lanz/xxxx is Lossless + obvious-POSTPROCESSING... you guys remember those Emulator filters, same concept.

You mean like adding scanlines etc. so make it look more like a CRT? I wouldn't put those in the same bag as a resize. A resize is trying to approximate what an "original" image would have looked like if it was at 2560x1440 while those Emulator filters are adding information that is not derived from the source at all.

I am not saying you should not use NN for yourself but trying to say a resize is the same as an effect filter is going too far.

If I take a 2560x1440 source, resize it to 720p, and then display it at 2560x1440 it looks more like the original using Jinc3 than NN. Jinc3 is a bit softer and NN has terrible aliasing.

Originally Posted by Fullmetal Encoder

With 4k screens coming I shudder to think what scaling up DVD content would look like. At this rate everyone will need two different screens to enjoy all of their content properly.

Why? What is worse about 480p/576p -> 2160p compared to 480p/576p -> 1080p?

Watching a DVD on a 70" screen I can understand some shudders but that is due to the size not the resolution, the extra resolution only helps.