Why SMF maps look so horrible

[smoth]

That's a lot of near-Moire.

I don't understand, it looks many times better than the old way.

[Beherith]

Ok, so heres my current version. It is close to the optimal solution for the mipmap level without texture strech compensation(yet)



Notice the great sharpness and the lack of moire patterns.

Please compare to other two screenshots and express your opinions.

Max detail:


current spring version:

Current state:
Stretch compensation will come with a spring startup time penalty (currently about 1.5 sec on my pc and dsd), since im convolving the heightmap with a sobel kernel to get stretch factors. I also plan on adding control based on current spring param texturelodbias.
So its a bit wip, but very close to done. Ingame fps penalty is negligible.

edit: done with stretch compensation as well, but i gotta run, will post screenies and patch laters.

[smoth]

I still like screenshot 190 more.

It looks much crisper.

[Argh]

Looks great, Beherith! Good balance there

[very_bad_soldier]

I still like screenshot 190 more.

It looks much crisper.

+1
195 looks a bit washed out.

[Beherith]

When the camera is moving you can see the moire patterns on the previous max detail screenie. Its because all mips are forced to max, which is extremely wasteful and not so pretty.

My last screen looks flawless in motion, because it uses the optimal sized mips for the job. I cant help it if the mips themselves are blurred and cause your altered perception of quality.

Proof that dsds mips are blurred can be seen by the fact that it has scanlines, since these can only appear when using filtered mips. I plan on rereleasing dsd with updated options and higher quality to show you.

Patch coming tomorrow :)

[smoth]

cool. thanks for the explanation.

[Forboding Angel]

When the camera is moving you can see the moire patterns on the previous max detail screenie. Its because all mips are forced to max, which is extremely wasteful and not so pretty.

My last screen looks flawless in motion, because it uses the optimal sized mips for the job. I cant help it if the mips themselves are blurred and cause your altered perception of quality.

Proof that dsds mips are blurred can be seen by the fact that it has scanlines, since these can only appear when using filtered mips. I plan on rereleasing dsd with updated options and higher quality to show you.

Patch coming tomorrow :)

If you're going to do that, would you be so kind as to retexture it?

[smoth]

If you're going to do that, would you be so kind as to retexture it?

that is coming after I move.

[Beherith]

Ok, its done and tweaked. Takes about a second on load to generate the sobel maps, but otherwise it works like a charm, compensates for extreme heights as well as just normal cliff streching. Provides optimally sized mipmaps.
My new mapconvs sinc filter greatly enhances the quality as well.

If a dev with commit access would test this out and maybe even commit it, I would be very happy :)

BFGroundTextures.cpp

(11.59 KiB) Downloaded 135 times

BFGroundTextures.h

(1.01 KiB) Downloaded 75 times

[Beherith]

[Forboding Angel]

behe, can you tell me, what needs to happen with mapconv to use your uber duper awesomestuff?

And could you link me? or is it not that simple?

[Argh]

Tested the latest commit. It's an improvement, and runs fine

[Beherith]

Forb, here is my mapconv: http://springrts.com/phpbb/viewtopic.ph ... 15#p399915

Thanks tobi for committing this patch!

[Tobi]

Auswaschbar beat me to it actually, although I had to remove that debugging line we spoke about cause he didn't see that

[Caydr]

Does this require maps to be re-released? Also, most of your images are missing and I'm curious what these guys are all hot and bothered for. Before/after shots would be really nice.

[Beherith]

Maps wont have to be rereleased, also, by nature it should reduce scanlines as well. Since scanlines are more visible on the smaller mip levels.
For some reason my host randomly eats some images, for the life of me I cant figure out why.