On Wed, Jan 28, 2009 at 12:17 PM, Andriy Fedorov <span dir="ltr"><<a href="mailto:fedorov@bwh.harvard.edu">fedorov@bwh.harvard.edu</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Daniel, Bill -- thanks for replies.<br>
<br>
I wonder, how are these multiple baselines generated? Are they<br>
collected from running the test on different platforms? </blockquote><div><br>yes.. floating precision implementations, ceil /floor/rnd implementations, compiler optimizations etc cause minor differences in the result, even when running the same algorithm, initialized with the same seed. We just check multiple baselines if they are deemend acceptable.<br>
<br></div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Or by<br>
adjusting rotation and translation components in all possible<br>
combinations by some (0.1?) tolerance, and applying the transform to<br>
the input? Seems like a daunting task in 3d...</blockquote><div><br>No <br></div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><br>
<br>