Shaders Operational

Post by **hellcatv** » Sun Sep 02, 2007 10:57 pm

chuck: that was only one of the bump maps---also your conversion technique must have some small bug in it... it doesn't look right in engine because the images must be rotated so they face northerly

Post by **chuck_starchaser** » Sun Sep 02, 2007 11:59 pm

hellcatv wrote:chuck: that was only one of the bump maps---also your conversion technique must have some small bug in it... it doesn't look right in engine because the images must be rotated so they face northerly

Fixed (I hope) back in the same link. Not sure what other bump maps you refer to; I only saw one in the link Phlogios posted.

Post by **loki1950** » Mon Sep 03, 2007 1:53 am

Everybody is so busy checking out the ship details nobody is looking at the scenery

so to make up for that here's a fun one

Where's Space Ghost

Enjoy the Choice

Post by **hellcatv** » Mon Sep 03, 2007 3:11 am

which system is that in...
probably more textures that need to be reverted until safemode has some idea of why certain textures appear to be white

Post by **chuck_starchaser** » Mon Sep 03, 2007 7:01 am

Rings could use a 1D, radial texture; unless you want them braided. What could never happen in a trillion years is having two ring systems at an angle.

Post by **ace123** » Mon Sep 03, 2007 10:04 am

Beyond that is the fact that part of the rings are showing up in front of the planet.

Post by **hellcatv** » Mon Sep 03, 2007 10:46 am

well that could be fixed by subdividing the ring mesh more---it's an option in the system file...

Post by **loki1950** » Mon Sep 03, 2007 2:42 pm

The system is Agea i was on my way to the War jump point and the rings caught my eye forgot to target it for max info sorry.

Enjoy the Choice

rigelan · Post by **rigelan** » Mon Sep 03, 2007 3:10 pm

I found one of those in Everett, I think. Pretty exciting! Lots of little issues in that one.

Phlogios · Post by **Phlogios** » Tue Sep 04, 2007 1:12 am

I have finished a bump map for the Ancestor ship. It is up on my ftp at: http://uku.hansaworld.net/~gustav/Ships/VS/bumpmaps/

I guess you don't have to post screenshots, because HellcatV is making a Mac build for me

, but if you want to (I know you just have to, klauss) please do. I'm not implying anything

Does anyone mind me making a bump for the Schroedinger?

Post by **loki1950** » Tue Sep 04, 2007 1:24 am

Does anyone mind me making a bump for the Schroedinger?

Well someone has to do it

and you are setup for it after all lots of other ships to do enough for all

Enjoy the Choice

Post by **chuck_starchaser** » Tue Sep 04, 2007 1:48 pm

Here you go:

Post by **klauss** » Tue Sep 04, 2007 3:40 pm

Just an idea, chuck - what about adding some low-frequency noise to the panels in the bumpmap?
I bet that would look a lot better than a flat bumpmap, and you can easily try it in the engine now

Just an idea of something that could be standardized, instead of flat bumpmaps.
BTW: low-frequency as in spots the size of panels, or comparable if smaller.

BTW: Daniel, after seeing this bumpmap, the new shaders do respect the standard

- so no change needed.

Post by **chuck_starchaser** » Tue Sep 04, 2007 4:23 pm

Good idea.
Here we go (hope it's not too much):

Try this one, too:

I compressed the angles by cranking up the blue channel and renormalizing.

Post by **klauss** » Tue Sep 04, 2007 4:58 pm

Cool - I'll try them when I get home today, however my idea was an even lower frequency. As in large scale panel shape irregularities - they look really nice, from my experience.
I would try myself but I can't build the gimp-normalmap plugin, my GTK is all messed up, and I'm a bit afraid to try to fix it right now (I tried once and messed up a lot of other stuff, and right now I need a functional system).

Post by **hellcatv** » Wed Sep 05, 2007 11:05 am

chuck: have you experimented with crazybump

http://www.crazybump.com/

it apparently makes bump maps from shader pictures or something

I can't get it running--maybe since you have so much trouble with everything else it could work

Post by **klauss** » Wed Sep 05, 2007 10:29 pm

I tried the bumpmap and the noise still has too much high-frequency content. IT looks like stone. I'd make it much smoother, very very smooth. Or, better yet, do it directly on the normalmap, and not the bumpmap - that way you'll get smoother results. The shader renormalizes after all, so you don't have to worry about those things like normalization and such.

Post by **chuck_starchaser** » Thu Sep 06, 2007 1:17 am

Try that. If it doesn't work, give me the mesh, and I'll load it into blender, make a high poly subsurf version, then make a normalmap from the two with xNormal. Not only I get excellent smoothing doing that, but in addition, subsurf creates some beneficial artifacts, little undulations here and there that look like soft wrinkles on the metal from stress. I can feed the bumpmap to xNormal at the same time, and it uses it too.

Post by **chuck_starchaser** » Thu Sep 06, 2007 2:30 pm

Here's an example of what using the xNormal method will do. I just re-did the Demon normalmap, after turning the UV layout so that the ship's nose points up in the textures. So, I exporte the mesh to a demon_low.obj file. Then I subdivided the mesh using sub-surf x4, and exported that as demon_high.obj. I fired up xNormal and fed it the two meshes and my bumpmap texture, 50% blending, and let it rip. Then in Gimp I loaded up the generated normalmap, doubled the contrast of the red and green channels, lowered the brighness of the blue channel by 25%, and used the nvidia plugin: Filters -> Map -> Normalmap -> Normalize Only, and VOILA:

http://deeplayer.com/temp/demon_normalmap.png

Check it out in detail; you'll see lots of features all over the place, in that normalmap. It nowhere "sits quiet"; except in the background, between UV map islands, of course.

When applied, not only it makes the guraud shading artifacts disappear, but it also adds some artifacts of subsurf that look like subtle wrinkles here and there, that make the mesh look more believable, if not realistic. This shot is from Blender's renderer:

Perfect place to spot the difference:
The exhausts at the back produced too many errors in xNormal, so I removed them from the meshes, and you can see the results.
The muzze parts (chalk white, a bit hot inside) look very polygonal in spite of smooth shading. That's what the rest of the ship would look like.

Can you spot a polygon anywhere else in the ship? No? That's what the xNormal method does for you; plus it adds all those beneficial artifacts...

Post by **chuck_starchaser** » Thu Sep 06, 2007 3:52 pm

Here's a bottom pic from Render Monkey with my current shader, --to ease back into the topic

Notice in particular the elipsoids at the front, how smooth they look. If you go back and look at the normalmap in the previous post (first link), you'll find a lot going on at those ellipsoids. At the current level of subdivision (which you can tell from their outlines) I would have had tons of guraud shading artifacts that the xNormal generated corrections smoothed over.

You can see how much simpler and boring the exhausts look:

The reason the rivets look funny, I reason, is the lack of specular self-occlusion.

Post by **klauss** » Thu Sep 06, 2007 10:17 pm

Chuck... did you try xNormal maps on the engine?
The shaders right now don't take a tangenspace, they make it up, and so baked normalmaps could be broke. Could, I'm not sure, and it'd be very important to find out.

Post by **hellcatv** » Thu Sep 06, 2007 10:21 pm

Ya chuck--it's definitely critical to be working with the engine and make sure all your theory translates into "fact"

as for the low framerate on your computer... does that occur with other openGL games...I've asked you to try Defcon, quake3...and I'll ask you to try vanilla Privateer Remake 1.2 as well (since you claimed that one used to work for you)

let us know if the framerate is acceptable for ANY of those games

My uninformed guess will be: "no"

Post by **chuck_starchaser** » Fri Sep 07, 2007 12:25 am

klauss wrote:Chuck... did you try xNormal maps on the engine?
The shaders right now don't take a tangenspace, they make it up, and so baked normalmaps could be broke. Could, I'm not sure, and it'd be very important to find out.

Is there something more to it, besides the nose of the ship pointing up in the texture?

@Hellcat: Okay, I just verified my framerate is as smooth as silk with PR1.2.
I'll try Quake 3 next...

Post by **chuck_starchaser** » Fri Sep 07, 2007 6:47 am

Okay, Defcon played just fine; in fact, I played it all evening..

Quake 3 Arena crashes. Even on a clean reboot, it crashes. That's after initializing the screen and leaving it so bright I can hardly see to reboot.
The message always is that the instruction at address x tried to access address x, but the memory there is not writable", paraphrasing. The address of the instruction is always the address it supposedly tried to write to. But each time it crashes it's a different address.

Post by **klauss** » Sat Sep 08, 2007 2:50 am

chuck_starchaser wrote:
klauss wrote:Chuck... did you try xNormal maps on the engine?
The shaders right now don't take a tangenspace, they make it up, and so baked normalmaps could be broke. Could, I'm not sure, and it'd be very important to find out.
Is there something more to it, besides the nose of the ship pointing up in the texture?

Ha! you bet!
Normalmaps ain't said to be in "tangentspace" vs "objectspace" for nothing. First of all, nose pointing up is quite inconsequential, if you ask me. When you create the bump, the effect of rotating is a swizzle on output colors, and it's the swizzle all that's important. The swizzle, IIRC, is to make red be left-right (since red=tangent, and the shader makes up tangents to be in the general direction of left), and green front-back (since green is binormal, and with a left-handed tangent, and an upwards normal, you get a front-facing binormal - three-finger rule). Ok... so, that's what the front goes up rule is for.
Now... you get the point now?
In case you didn't notice, let me get your attention to it.

I wrote:(since red=tangent, and the shader makes up tangents to be in the general direction of left), and green front-back (since green is binormal, and with a left-handed tangent, and an upwards normal, you get a front-facing binormal - three-finger rule)

See?
What happens on the bottom of the ship? The normal goes down.
Three-finger rule again, and where does the binormal point to?
Oops - back.
But the tangent still goes towards the left (it's hardcoded that way). So... what? Well, when you interpolate tangentspace through a face which rotates the normal, tangenspace also rotates in a way incompatible with what bakers expect (bakers expect the tangent pointing towards the +s direction - the direction in which s-coordinates increase, which in this case would be left, not right). So, when your baker's tangentspace and the shader's tangentspace don't match, you get - inconsistencies. Some big, some small. Or, maybe, if you're lucky in the way you textured your ship, maybe none noticeable. That's what I'd like to find out.