Suggestion for alternative handling of interlaced frames (WinUAE)

[Photon]

Looks fine here with moving object, can't see any difference at all with/without "remove interlace artifacts". In both cases, simply artifact-free pictures with twice the vertical resolution.

It runs at 25fps tho, that's probably it.

Any tips for some game or such that has objects moving in interlace at 50 fps!?

[Maren]

Originally Posted by Photon

Looks fine here with moving object, can't see any difference at all with/without "remove interlace artifacts". In both cases, simply artifact-free pictures with twice the vertical resolution.

That's weird. The problem was confirmed by Toni himself, so it's definitely there. If you set a screenmode with a vertical resolution not big enough to handle a whole weaved frame, then fields would be displayed consecutively instead of progressively and you would get no artifacts at all, but then you'd be seeing only half the vertical resolution, and I presume that's what happened to you.

Originally Posted by Photon

It runs at 25fps tho, that's probably it.

Actually the game was not designed to be displayed on progressive screens, so technically it runs at 50 fields per second.

Originally Posted by Photon

Any tips for some game or such that has objects moving in interlace at 50 fps!?

Use hardware/drivers capable of outputting video interlacedly or wait for Toni to come up with a fix

[john4p]

Originally Posted by Photon

Any tips for some game or such that has objects moving in interlace at 50 fps!?

Slam Tilt in "hires"-mode.

[Photon]

Misperceive.

Looks fine = MY moving object displays fine, buffer contents update at 25 fps.

Tips for games = tip me off to a game that uses interlace, so I can try it in WinUAE!

I'm still using 1.6.1, if that's a clue.

[Toni Wilen]

Try any demo that has interlace scroller (I don't remember any names..) -> really ugly artifacts. Pinball Illusions in lace mode is also good test game.

[StingRay]

Originally Posted by Toni Wilen

Try any demo that has interlace scroller (I don't remember any names..)

This Agile intro should be a good test candidate.

[Photon]

Yes indeed Stingray

Test results: All modes have direct3d bilinear 16 bit filter and hires (normal) mode selected.

For Windowed:
Line mode:
- Normal: vectors have no artifacts, hireslace scroller no artifacts, but is jumping 1/2 line.
- double: artifacts on both
- scanlines: artifacts on both

Fullscreen+Vsync:
Line mode:
- Normal: perfect result except scroller is jumping 1/2 line (funny... I seemed to have had that perfect one time, but now I can't get it back. Maybe I simply didn't "look for it" before...
- scanlines: artifacts on both
- double: artifacts on both

For all these modes, the results are the same with or without "reduce interlace artifacts".


Toni: I noticed a tiny 'feature' that I had noticed before... when selecting some display settings and check what they look like, you get faint almost-black-on-black border flicker. I noticed now that simply pressing F12 and OK removes it!


After these tests I made a config that I like:

[Photon]

I should add that I had "autostretch" on the Direct3D filter in the tests, and in the submitted config I had 2x horizontally and 4x vertically instead.

[Maren]

At this point I'm wondering, how is WinUAE supposed to handle combing? On progressive monitors, interlaced Amiga screens behave a bit like video from camcoders, and the only way to do proper deinterlacing after weaving this type of video is introducing ghosting, which doesn't look as bad as combing but you're basically replacing an artifact with another.

Would WinUAE just detect the combed areas and remove them all together?

Thanks.

[Photon]

Have you tried it in WinUAE? I don't care how he does it, it looks perfect to me

Tested the Agile demo on my A600 portable, and the Joytech screen really is wonderful! It looks even better than in WinUAE! (Well, only slightly. Since it translates to lowres vertically, the jumping line is missing.)

I love how the Joytech takes two interlaced lines, and translates them to 1 lowres line, and makes half-pixel-"areas" half-transparent It also preserves hires, so the scroller in the demo looks ace! Hires with vertical antialiasing.

The vectors below are completely lowres on it, though - I think it doesn't like the hireslace->loreslace mode change in the copper, if the whole screen had been hireslace it would have been antialiased vectors

Not a single pixel flickering. I love my little screen

[Maren]

Originally Posted by Photon

Have you tried it in WinUAE?

Of course, but since it doesn't work correctly here, I don't know how it's supposed to look like.

[LocalH]

Originally Posted by Maren

and the only way to do proper deinterlacing after weaving this type of video is introducing ghosting

Double-framerate bob is the simplest way to handle such material. Of course, there are more advanced deinterlacers that work at both single-framerate and double-framerate, but bobbing is computationally cheap and (on a CRT) even slightly emulates the interlace flicker but not nearly as much as on your average 1084.

[Maren]

Originally Posted by LocalH

Double-framerate bob is the simplest way to handle such material. Of course, there are more advanced deinterlacers that work at both single-framerate and double-framerate, but bobbing is computationally cheap and (on a CRT) even slightly emulates the interlace flicker but not nearly as much as on your average 1084.

http://eab.abime.net/showpost.php?p=587721&postcount=15

[Toni Wilen]

Note that I only care about pixel-perfect image, I don't care about video algorithms.

[Maren]

Originally Posted by Toni Wilen

Note that I only care about pixel-perfect image, I don't care about video algorithms.

There really isn't much of an algorithm in bob deinterlacing, it's just fields being upscaled to twice their vertical resolution, then bobbed up and down, which looks horrible for low-res pixel art anyways, and that's where the method I suggested would come in handy

[Toni Wilen]

The point: there is full non-interlaced image in Amiga memory, in other words we (nearly) always have the original pixel perfect data.

We only need to find a way to automatically detect how it is rendered in interlace (by analyzing copper list etc..) and then draw it (current method simply uses previous and current frame's bpl pointers and draws both odd and even line at the same time) in 50fps 512p without quality loss or any image processing.

[Maren]

The problem with interlaced amiga games is that we are not dealing with a 100% progressive source. The backgrounds are static, so no problem there, but the movement of the foreground objects is generated in real time, which means we can't do any sort of prediction thus an ivtc-like method for deinterlacing amiga frames would definitely be out of the question if we were to play by the same rules that apply to regular video.

So what's the approach here? removal or prevention of deinterlacing artifacts?

[Toni Wilen]

Originally Posted by Maren

The problem with interlaced amiga games is that we are not dealing with a 100% progressive source. The backgrounds are static, so no problem there, but the movement of the foreground objects is generated in real time, which means we can't do any sort of prediction thus an ivtc-like method for deinterlacing amiga frames would definitely be out of the question if we were to play by the same rules that apply to regular video.

So what's the approach here? removal or prevention of deinterlacing artifacts?

There is no approach, I am not interested in this case

EDIT: I missread your reply. Many games have progressive source, even if interlaced. This is the situation I support. (game in OP is also progressive, current simple method does not work because it has multiple buffers)

[Maren]

I just realized I know less than jacks*** about buffers or how programmers can deal with de-interlacing artifacts in ways extraneous to the traditional methods favored by video technicians, so I won't comment on that , still, would the same potential fix also address the combing in Workbench 1.3?

[Toni Wilen]

Originally Posted by Maren

I just realized I know less than jacks*** about buffers or how programmers can deal with de-interlacing artifacts in ways extraneous to the traditional methods favored by video technicians, so I won't comment on that , still, would the same potential fix also address the combing in Workbench 1.3?

It already does but it detected KS 1.x interlace modes as incompatible, fixed in latest beta.