Seams with mirrored parts tweaking

I have a problem with the tangentspace bumpmapping of mirrored parts. The problem is NOT the mirrored parts in itself its where the mirrored parts meet the unmirrored ones. This happens with things like a face that’s only half skinned. I’ve tried the following things but all have some visible glitch. Do you guys have done more tweaking with this case? Any suggestions I could try?

These are what I already tried

  1. Just using 2 bases per vertex one for each side of the mirrored part http://tenebrae.sourceforge.net/beta/tangent_1.JPG

  2. Using the averaged space of the 2 bases from (1) This obviously gives a degenerate s pace as the vectors cancel each other out http://tenebrae.sourceforge.net/beta/tangent_2.JPG

  3. Using only one space (the unmirrored one) this gives a seam as the space “flips” over the first row of mirrored tris. http://tenebrae.sourceforge.net/beta/tangent_3.JPG

Charles

I can’t tell from the pics, but method 1 should work.
What’s the problem with it? Isn’t it suppose to look like a clown mask?

No it’s not supposed to look like a clown… it’s supposed to be a scientist (when he will be finshed)

I just tried the progam and i very impressed… altough it is quite slow on this computer (duron 900/gf2 mx400). but it is a prefect example of how great bump mapping looks.

Jan

Yeah, Willi is a great artist and he’s getting really good with this real-time stuff too. He’s the one that made those materials, right?

I think technique 1 should work too, but it looks like the tangent spaces aren’t properly smoothed on the mirror line. I think the two tangent spaces should be identical with only the tangent (or binormal) vector flipped on the seam.

Another thing causing the seam can be a specular approximation, which might not behave properly on seams. At least mine creates seams when the tangent space isn’t continuous, but I think my specular is a bit unorthodox anyways so I don’t know whether this is a common problem or not.

-Ilkka

Strange I’ll do some more experiments at least I know the direction I should follow now
It will always be somewhat hacky I guess I checked D3 even there you have minor seams where the parts meet. Maybe object space bumpmapping would be better for skinned meshes…
Well the model is done by Pestilence with a polybumped bumpmap. (With ORB)
The wall textures are indeed Willi’s

Charles