Normal Maps, Antialiasing and render engines

Jhotun

Hi people

I would like to know which antialiasing method gives the best result when baking normal maps. Also, I have experienced different results using mental ray and scanline for the baking process (same model, same cage), so I´d like to know which method is being used mostly.
Another thing that I´d like to know is about normal map size. Do you create a 2048x2048 and reduce later to fit the model requeriments or simply extract it at the final resolution?

Thanks a lot!

CrazyButcher

about normal map size: simply rescaling is not a good idea, since rescale filters do not take into account that your pixels must represent a normalized vector. Likely there is filters that allow to renormalize them after scale operations.

Kevin Albers

You can renormalize the maps using the Nvidia photoshop filter. I typically use the scanline renderer when baking, and bake the normal maps out at a higher resolution than final size.

Eric Chadwick

You get more accurate results by increasing the AA setting (# of rays cast), but it takes I think exponentially longer to render (or at least it seems so!).

Like Kevin, we render bigger (2048 for 1024 output) with lowest AA setting (1 in Kaldera), then scale down in Photoshop, then use the Nvidia filter to renormalize. Nvidia also has a nice tool for creating good mip maps for the normal map.

The difference between the two methods is pretty minor. But best way to see is to try yourself.

Jhotun

Renormalize after the rescale (with the nvidia tool) is enough?

EarthQuake

We just render at true res with multi-sampling, since we use quite a few 2048x2048 maps and render out GI light bakes at the same time rendering out double size is really not a valid option.