16bit vs 8bit vs Raw source images

Using 16 Bit uncompressed tiffs works in RC but gets interpreted in view and construction blown out colorwise, so does 16 bit depth in source photos yield better/more accurate results for alignment and reconstruction at all?

Will downconverting to 8bit lower the achievable quality?

Raws are supported but dont seem to yield nicer results than 16 bit so far.

 

 

Additionally we found that the initial feature points detection before actual alignment process starts finds about double the amount of #pts for an 16bit compared to 8bit. Resulting in lower median and mean error rates in the alignment report as well.

Also ultra detector sensitivity makes a difference.