[Bilbo]

1. I have two normal maps for a lowpoly object- the first was generated from a highpoly object, the second was generated from a height (bump) map. What i want is to combine the two maps together- i.e overlay the second map on top of the first- has anyone had any experience with that? will using the Photoshop Overlay mode do the trick ? any other ideas?

2. Does anyone know of a free openGL program for previewing Normal-mapped models ? I'm looking for a compact piece of software, no need for any advanced features or tools.

(thanks)

[Frost]

Hi Bilbo.

The easiest way to do this is to 50% blend the both together (then renormalize). This will result in a non-normalized normal map. You can use free tools like nVidia's nvTools NormalMapFilter (on this page, which may require the "Put these dlls in windows\system (325 KB)" zip file as well) to do this, by selecting "normalize only" from the panel. The tool also lets you preview the normal map on a 3D quad interactively.

These tools are fun in that you can paint some normal maps manually if you know what you are doing, renormalizing with the tool above.

If you do this, you might have to rescale as well... strangely, OVERLAY is not working exactly the way I expected in PS... I'm investigating...


Have fun, and good luck.
Frost

[Bilbo]

Thanks Frost! yeah, i found out the hard way that Overlay doesn't do the trick... i didn't know i could renormalize things with this nvidia filter, you've been of great service.

- EDIT - did some testing - seems like by blending the two normal maps together i can't help but affecting the original map's intensity - and since we're usually dealing with maps generated automatically from 3d objects, it lessens the credibility of the final effect. Is this what you meant by "rescaling" ? how is it done? how can i get it to fit the original scale again?
I noticed that ATI's normalmap software does a pretty good job at combining a normal map and a bump map , but it seems to only work with bump maps on the highpoly object , which is a pain because otherwise the highpoly object doesn't need uvmapping at all.
sorry for babbling.

[Rychan]

I'm skeptical of using image editing software to combine normal maps. How are your normal maps encoded in the image? Each R, G, B color channel representing the normal in x, y, z? each pixel scaled to some constant intensity (normalized)?

I suppose if you were adding 255 red (x vector) to 255 green (y vector) you would want to get 255* sqrt(2)/2 red and green?

I'm pretty sure that any single overlay mode in photoshop will not be this smart, as they tend to operate on individual channels
(http://www.pegtop.net/delphi/blendmodes/ is a good reference)

you might get reasonable results by just blending them in normal mode (the top one at 50%), then putting a luminosity layer above both of them to try and force some kind of intensity equalization.

It would be pretty trivial to write software to do this correctly, though.

[Frost]

Bilbo: "it lessens the credibility of the final effect. Is this what you meant by "rescaling"" -- yeah, that's exactly what I meant. I'll think about it some more... maybe there's something to be done about this... I first thought that the "Overlay" filter was an "add_bias" operation, but, now I'm not sure anymore... some tests are in order. (edit: hm, no, it's not. Stupid useless mathematically braindead filters...)

Rychan: Thanks for the link... I hope he did his homework right... just his notation makes me itch.

"It would be pretty trivial to write software to do this correctly, though." -- I think so too...

"Each R, G, B color channel representing the normal in x, y, z? each pixel scaled to some constant intensity (normalized)? " -- Yes, each color channel in RGB represents the XYZ coordinate endpoints which describes a vector from the origin (XYZ 0,0,0) to that XYZ coordinate. The RGB values complement each other so that the length of the vector is always 1 (unit length) -- a "normal". These XYZ coordinates can range from -1 to +1 on either of the XYZ coordinate axis, so they are saved in a signed byte format per color component (RGB)... thus having -1 coordinate values represented as 0, and +1 coordinate values represented as 255, where the neutral (center) value of 0 is represented as 127 (127.5, but uhm, we round off to something like 127 or 128, depending on the software =). But you seem like you know all that already...