Question about Normalmap Mirroring
polycounter lvl 17
I've been experimenting lately with normalmap mirroring.
1. I've been able to use duplicate normalmaps with no problems in 3ds max.
2. Any flipped uv's don't get their normalmap flipped properly in 3dsmax, but a lot of engines support this function - I'm curious is there a way to make 3ds max support this ?
3. Is it possible to mirror geometry which is connected in the final lowpoly ?? I have an example below.
I've tried changing the smoothgroups in the two sections, and also in another test used the same smoothgroup.
1. I've been able to use duplicate normalmaps with no problems in 3ds max.
2. Any flipped uv's don't get their normalmap flipped properly in 3dsmax, but a lot of engines support this function - I'm curious is there a way to make 3ds max support this ?
3. Is it possible to mirror geometry which is connected in the final lowpoly ?? I have an example below.
I've tried changing the smoothgroups in the two sections, and also in another test used the same smoothgroup.
Replies
A single smoothing group should work, however Max's in-viewport preview of normal maps isn't 100% accurate, you'd probably get better results if you viewed the model and normalmaps in a game engine (or maybe try Jogshy's XNormal program).
I don't think there's any way to get Max's viewport to 100% accurately represent normals the way a game engine would at the moment, for some reason.
I've tried loading this model in different engines, such as xNormal and Doom3 - and there seems to be a line crossing the section in both of them.
I've also checked the projection, and fitted the borders to x=0, this improved things but the problem is still very much there.
I know my best bet would be when I get access to the engine this gun is made for, nevertheless I have a feeling this problem isn't because of engine compatibility.
I think this is caused because on bake time, only one half of the geometry is present, and the edge normals are different from when both pieces are connected.
I'll do some more tests ...
That was exactly why the problem occured, and was able to fix it :
1. the initial projection borders weren't at x=0.
2. using the same smoothing group for both sides, while baking only one side means the borders will have different normals and produce different results. (so I just assigned different groups for both of them, since this was the simplest route)