Artur J. Żarek

Hi Tamte and thanks for your reply. Your second paragraph pretty much describes my goal. I have two geometrical representations of the same object: one subdivided, one not. Both have the same UV seams and UV positions of seam vertices. What I'd like to do, is to investigate and assess the resulting quality of skipping normal map baking, and instead, directly applying rasterized normals of the highpoly object to its lowpoly counterpart, thanks to texture processing capabilities of COPs. Without user interaction or creation of additional files (which texture baking requires). I'd like everything to happen in COPs, except of maybe TBN calculation of target lowpoly object, which I believe is required for the conversion and needs to be calculated in SOPs (if I'm not mistaken, "Tangent" convenience SOP based on PolyFrame SOP does it out-of-the-box).

So, what I'm looking for, is the proper algorithm/pseudocode/step-by-step instructions for conversion of rasterized WSNM of high poly geometry to TSNM of the low poly one, or (and ideally) a robust COPs recipe for this operation. I've seen so many formulas and solutions on the interwebs, that I'm not sure which one is the correct one anymore.

What are the steps required for converting the result of rasterization of N vertex attribute, which I believe is (and please do correct me if I'm wrong) a signed world space normal map, into a proper offset tangent space normal map? Can this be done with currently available COPs, or does it require a custom implementation in OpenCL?

Gosh, I guess old habits die hard.
Thanks for pointing this error out. I too find it surprising that the compiler didn't detect it.