with nt are the new texture coordinates, t the normal ones, l the transform light-view vector in tangent space and im_w and im_h the texture size.
The main problems are im_w and im_h which I think are not really consistent, but didn’t found out any other better solution.
I can see a little displacement between normal texcoords and shifted ones when I alternatively
display the object with both tex coords.
Also, I’m not totally sure on the steps, here is what I actually do:
Render the object normally.
Render the object with shifted texcoords with blend func set to (one,one).
Render again the object with normal texcoords with blend func set to (dst_color, src_color)
Render again the object with normal texcoords but with gouraud shading only.
I don’t know why you have two ordinary passes in there unless you’re modulating by texture, but you say you just gouraud shade… on top of your bump map result?
I try to do it exactly as the help stipples it. I do not use multitextures for the moment. So I render exactly the same object several times using blending but without using the accum buffer.
And yes I gouraud shade on the top of the previous passes. When I simply avoid that last pass, I can’t notice any difference at all.
I also read the document you gave in link, but didn’t really helped me for the moment. I don’t really understand what derivative to use, in fact I don’t see regarding what dimension the derivative is made (something like dx/dt or du/dx would have had helped me).
I tried to create emboss bump mapping, but I think I failed because the results are not convincing.
Of course not. It’s emboss bump mapping; it almost never looks convincing, even when correctly implemented.
The emboss technique is a poor approximation of per-pixel lighting. Unless you are just investigating the technique, or your target hardware is pre-GeForce1 hardware, it is better to use other bump mapping techniques.
Go look at my example, if you can’t get it looking as good as that you’re doing something wrong. I show the results after each pass you should be seeing too you can do this also as a means of verifying correctness.
Don’t forget the glBlendFUnc option to boost brightness in that last pass because you have to avoid the clamp in the second pass.
What to understand when saying object space coordinate ? I was beleiving it’s the the space of the object when it doesn’t have been transformed at all (no translation, no rotation…). But I could have read somewhere that I must do as if all vertices of the object where center to the origin.