Created on 02.10.2007 - 20:06

I was wondering... Is there a way for the 2xSaI shader work according to the level strength you put on the shader?
While I'm waiting on an answer for that question here is something you might enjoy.

This is a nice result I had when testing an idea on Pete's OpenGL2 GPU Driver with a 2xSaI shader (Credits to the 2xSaI shader go to Kreed, guest(r) & VerGreeneyes on their awesome work on this shader).

Anyhow, the game runs at 320x224 during gameplay, and the 2xSaI shader doesn't get the best results when set on "Internal X/Y resolution: [1: High]" So I tried incresing the size 50% (480x336). I also have the Full "Screen filtering: [Enabled]", "Texture filtering: [6: Extended]", and forced my Nvidia Driver to 8xS AA. The first image is No 2xSaI, No Filtering & No AA for comparison. Unfortunately this only works in windowed mode...






If you think these look too blurry just force your Video Card Driver to No AA and it will look sharper like this:






Emulator: ePSXe v.1.6.0
OGL2 Shader: 2xSaI
Nvidia Driver: 8xS AA

Plugin: Pete's OpenGL2 Driver 2.8
Author: Pete Bernert
Card vendor: NVIDIA Corporation
GFX card: GeForce 7600 GT/AGP/SSE/3DNOW!

Resolution/Color:
- 480x336 Window mode
- Internal X resolution: 1
- Internal Y resolution: 1
- Keep psx aspect ratio:
- No render-to-texture: Î
˛

- Filtering: 6
- Hi-Res textures: 0
- TexWin pixel shader: off
- VRam size: 0 MBytes

Framerate:
- FPS limitation: on
- Frame skipping: off
- FPS limit: Auto

Compatibility:
- Offscreen drawing: 0
- Framebuffer effects: 3
- Framebuffer uploads: 2

Misc:
- Scanlines: off
- Mdec filter: on
- Screen filtering: on
- Shader effects: 5/1
- Flicker-fix border size: 0
- GF4/XP crash fix: off
- Game fixes: off [00000000]

Hope ya liked them!

[Dieser Beitrag wurde am 10.10.2007 - 18:41 von GreenImp aktualisiert]

Created on 22.11.2007 - 23:40

Wanted to let you guys know, I've been working on this shader for the past week or so (I've also had the flu, so progress has been kinda slow) and I managed to speed it up significantly a while ago. I'm trying to write a little something that will brute-force simplify the conditionals involved, but it's a pain in the ass. Anyway, once I either finish this or give up on it, I'll post the speeded up version I've already got.

Created on 23.11.2007 - 10:01

cool. looking forward to it. hope you get well.

Created on 23.11.2007 - 17:23

Thanks; progress has been steady but slow (I hate iterative backtracking ...). Still, I hope to finish something tonight (though don't expect too much)

Edit: looks like it's not getting done tonight. I'm getting some output, but it's kinda messed up and the program crashes ...

Edit2: by the way, GreenImp, to answer your question at the top of the page: no, not really. The 2xSaI filter is an 'all or nothing' kinda filter. That's how it stays so sharp. The only way to make it dependent on the shader strength would be to blend the result with the original colour, but I dunno if that would look good.

[Dieser Beitrag wurde am 24.11.2007 - 01:00 von VerGreeneyes aktualisiert]

Created on 25.11.2007 - 09:54

Well, I've got it working. Now it's just a matter of time - and figuring out how much time! Might take a day, might take 500 billion years ... There are some things I could do to speed it up, and theoretically I could even try distributed computing (though I doubt anyone would lend their processing power to such a useless project)

Created on 25.11.2007 - 18:02

Well, i got an idea how to improove image quality.

Example:

You take the if (c4 == c8)equality situation.
Then you select the threshold th, like 5 singletons.

If p01 is examined regarding the upper condition, the final p01 could be computed this way (code is symbolic):

if (dif(C4,C8) less_equal th) p01 = mix(p01(C4==C8),p01(C4!=C8),dif(C4,C8)/th);
else p01 = p01(C4!=C8);

[Dieser Beitrag wurde am 25.11.2007 - 18:11 von guest aktualisiert]

Created on 26.11.2007 - 15:58

Your idea sounds good, guest, but there's one thing we should remember if we try this: as things stand, we're using a dot product that gives preference to Red, then Green, and only then Blue.. with range testing we'll have to use something like distance(C0,C4)<[some constant] in place of the current c0==c4

As a status update, the last version of my program would've taken infinitely long to complete :P I've figured out how to do it much more quickly, I think, but it'll take a rewrite.

[Dieser Beitrag wurde am 26.11.2007 - 16:01 von VerGreeneyes aktualisiert]

Created on 26.11.2007 - 16:14

This above example is just fine.
In fact one can't do much better if you want to realize 'smooth pattern transitions' with the actual 2xSaI code.

Linear metric is very suitable for this and you have to take a kind of color metrics if you want to realize smooth pattern transitions.

Created on 26.11.2007 - 17:09

I should've looked more closely at your example code, my apologies.

Created on 26.11.2007 - 17:10

I should've looked more closely at your code, my apologies.