Eros

[chris]

The NEAR spacecraft's laser rangefinder performed millions of observations of the asteroid Eros which were later compiled into a very detailed shape model. The finest resolution available model has samples spaced 1/8 degree apart. I'd always wanted to take this data and produce an insanely detailed model of Eros, and last night I finally got around to doing it. I ended up using the 1/4 degree data--the extra samples just seemed like overkill. Here's the result, first with a Lambertian surface:



And again with a Lunar-Lambert model, with a weight of 0.3 (just a guess):



One more image:



The models themselves are available here:

Quarter degree: http://www.celestiaproject.net/~claurel/celest ... 0.cmod.zip
Half degree: http://www.celestiaproject.net/~claurel/celest ... 0.cmod.zip

The half degree model looks very good; the quarter degree model is a bit too large to be practical (though I suspect that this may be because the model data isn't going into local video memory; the graphics chip should be able to handle the large triangle count.)

I eventually would like to use a lower resolution mesh and put the extra detail into a normal map.

--Chris

[Dollan]

So... the average user with an average video card (okay, less than average with only 30mb of memory to it) should use the half degree?

At any rate, this is a very cool piece of work! It certainly makes a great showcase.

...John...

[cartrite]

You could also divide the smaller model into 4 sections ( or more ) and assign each a part of a full normalmap created from all the data. I've done similar experiments with mars and find that the results look better with a normalmap and model that has a little detail to cast the shadows on as apposed to a highly detailed model that just about fits into memory with no normalmap. The trick is to find the right balance. For me and my setup the file size of the model and frame rates are about the same due to the tangents and extra files needed to use normalmaps. Using normalmaps look better though.

i just checked out the 1440 model and the detail is incredible. How did you create this, with modeling software or a custom program?

The 720 model has incredible detail too. It's hard to tell them apart except the better frame rate for the 720 model. The details look almost the same.

cartrite

[ElChristou]

Terrific!!

it would be cool to do a model with the larger features, the rest as you said via a normal map... Any way to generate a 3DS to edit the model?

[Cham]

There are some holes through the poles, and a kind of pinch effect around them. I think a simpler model with bump mapping (or normal map) could be better. Chris, any chance we could see bumpmapping on non-spherical models (asteroids), in the next official release of Celestia ?

[chris]

There are some holes through the poles, and a kind of pinch effect around them. I think a simpler model with bump mapping (or normal map) could be better. Chris, any chance we could see bumpmapping on non-spherical models (asteroids), in the next official release of Celestia ?

I've already got bump mapping and specular maps working for non-spherical models. I checked the code into CVS a few weeks ago. It only works with OpenGL 2.0, however.

I will improve the model by eliminating most of the triangles near the poles and adding tangents for normal mapping.

--Chris

[chris]

8O Terrific!!

it would be cool to do a model with the larger features, the rest as you said via a normal map... Any way to generate a 3DS to edit the model?

I'll see what I can do . . .

--Chris

[chris]

You could also divide the smaller model into 4 sections ( or more ) and assign each a part of a full normalmap created from all the data. I've done similar experiments with mars and find that the results look better with a normalmap and model that has a little detail to cast the shadows on as apposed to a highly detailed model that just about fits into memory with no normalmap. The trick is to find the right balance. For me and my setup the file size of the model and frame rates are about the same due to the tangents and extra files needed to use normalmaps. Using normalmaps look better though.

Splitting the model should be unnecessary . . . There's only enough data for a 2880x1440 normal map. A lot of graphics cards can handle as big as 4k x 4k. And for those that can't, you won't lose that much detail with a 2048x1024 version.

i just checked out the 1440 model and the detail is incredible. How did you create this, with modeling software or a custom program?

I converted the data from the NEAR probe into a 3D model with some quick and dirty custom software.

--Chris

[jestr]

If you were to make 3ds versions out of these models,they would need to be chopped up into many pieces.As far as I know the limit for a 3ds object is 65535 faces,whereas the 1440 mesh has over 2000000 faces and the 720 mesh has nearly 517000 faces.Splitting them up will mean many nasty seams unless these can be erradicated when you convert back to cmod format.I can give these models texture coordinates but I dont know if I can convert it to cmod again without splitting the model up as mentioned above,Jestr

[chris]

Here's the eros model 2 degrees with a 2880x1440 normal map applied:



The lighting uses the Lunar-Lambert model, and this image happens to be from a test solar system with two suns.

And just for fun, here it is again with some specular highlights to show off the bump mapping:



The actual normal map (created with a shiny new set of normal map building tools) is here:

http://www.celestiaproject.net/~claurel/celest ... s/eros.png

I should mention that there are few things that I know are incorrect about this normal map. The tool I was using is specialized for producing normal maps on smooth spheres; for the Eros map, I really should be accounting for the underlying geometry of the object, eliminating the part of the normal that's can be accounted for in the mesh.

--Chris

[Chuft-Captain]

Sorry, but I'm going to ask a basic question because things don't seem to be functioning as I expected, so maybe I've done something wrong.

I created a new directory "eros-720.cmod" in Celestia\Extras\ .
In this I created sub-directories: models\, textures\, and textures\hires\.
I also created an SSC:

Code: Select all

Modify "Eros" "Sol"
{
   Mesh   "eros-720.cmod"
   NormalMap "eros.*"
}


I put the SSC in Celestia\Extras\eros-720.cmod\, the model eros-720.cmod in Celestia\Extras\eros-720.cmod\models\, and the normalmap eros.png in Celestia\Extras\eros-720.cmod\textures\hires\.
I've made NO changes to Celestia\data\asteroids.ssc:

Code: Select all

.
.
.
"Eros" "Sol"
{
   Class "asteroid"
   Mesh   "eros.cmod"
   Texture "eros.*"
   # NormalMap "erosbump2k.jpg"
   # Color sampled from true color photo taken by NEAR
.
.
.

When I view Eros I see only subtle differences between LORES and MEDRES modes, and no difference at all between MEDRES and HIRES.
I was expecting more startling results from this normalmap (it is 7MB after all). This makes me think I have either installed something wrong, and as a result the normalmap is not being used at all, or it's my hardware, or my expectations are too high.
Can anyone offer any advice?
(Celestia 1.4.1, Windows XP on laptop with integrated graphics. )

Here is what I see:

LOWRES

MEDRES

HIRES


Thanks in advance
CC

[cartrite]

I suspect that you may have the wrong model. If you are using the same eros-720.cmod file I downloaded yesterday it has neither texoords or tangents.

[chris]

I suspect that you may have the wrong model. If you are using the same eros-720.cmod file I downloaded yesterday it has neither texoords or tangents.

Correct . . . here's a much less detailed Eros model that does have tangents:

http://www.celestiaproject.net/~claurel/celest ... s-tan.cmod

Personally, I think a bit of extra geometric detail would be nice, but it still looks good with the normal map. Also, the normal map is a non-power of two size; if your graphics card doesn't support that, you'll have to resize it (though rescaling it will likely degrade the quality because the normals will become de-normalized.)

--Chris

[Chuft-Captain]

Personally, I think a bit of extra geometric detail would be nice, but it still looks good with the normal map. Also, the normal map is a non-power of two size; if your graphics card doesn't support that, you'll have to resize it (though rescaling it will likely degrade the quality because the normals will become de-normalized.)

Thanks guys for the very swift replies,
I agree with you Chris. Although, FPS performance of this smaller model is much better for people with poor graphics capabiliies, I think perhaps a geometric detail level somewhere between the 2 extremes would be very nice.
I resized the NMap to 2K, so I suspect I may not have as nice a result as you have Chris with your NMap which hasn't been messed with.
Here's what I see now (Render->Ambient Light=Medium):



This one's a CelURL so (if you have time to post an image), I'd be interested to see what your setup produces from this viewpoint (with the eros-tan.cmod and a correct Nmap)


Cheers
CC

[chris]

I resized the NMap to 2K, so I suspect I may not have as nice a result as you have Chris with your NMap which hasn't been messed with.
Here's what I see now (Render->Ambient Light=Medium):

Hmm . . . in you screen shots I don't see much extra detail from the normal map. Have you compiled Celestia from CVS recently? Normal maps don't work with models in Celestia 1.4.1.

--Chris

[Chuft-Captain]

Hmm . . . in you screen shots I don't see much extra detail from the normal map. Have you compiled Celestia from CVS recently? Normal maps don't work with models in Celestia 1.4.1.

Damn! I had read a number of posts which lead me to believe that people were using Nmaps as far back as Oct 2005, so I thought that it must work in 1.4.1

I'm interested in seeing the difference, but I'm not set up to compile, so I guess I'll just have to wait until the next release to see the effects of Nmaps.
....unless of course someone with the latest CVS version would be kind enough to reproduce and post a screenshot of THIS URL for me.

[phoenix]

I'm interested in seeing the difference, but I'm not set up to compile, so I guess I'll just have to wait until the next release to see the effects of Nmaps.

or you just try this one

[selden]

NormalMaps have worked on Celestia's planets for some time. Up until now, only emissive and opacity maps have worked on Mesh objects.

[Chuft-Captain]

NormalMaps have worked on Celestia's planets for some time. Up until now, only emissive and opacity maps have worked on Mesh objects.

What about SpecularMaps? I think that I've read in a post somewhere that SpecularMaps only work in OpenGL 2.0 path, or with certain high-end cards... Can't remember exactly. Is this right? Can anyone briefly summarise what's required to use these.

Thanks in advance.

[Chuft-Captain]

I'm interested in seeing the difference, but I'm not set up to compile, so I guess I'll just have to wait until the next release to see the effects of Nmaps.

or you just try this one

Hi Phoenix,

Can you explain what this executable is. It's too small a download (1.5mb) to be a full Celestia build isn't it?

Thanks
CC