[Little case] reconstruct tangent from point cloud

We are engaged in constructing a tangent in the cloud at the point where the first bake has a good St. It is extremely simple to say. It generates an auto-scaling structure similar to a board layout based on the number of points, and then searches within each grid. As follows,

Because the sample labeled as best is closest to vec2 (), that is, to ensure the continuity in one direction, the calculation result is the dpdns of the current sample point (represented by a pentagram. In the same way, we can also get dpdt. Of course, we can also manually calculate these two components in RSL.

I ran 3 seconds on the t2250 + 2G machine in my notebook, and ran 1 second on the e6550 + 2G desktop. The speed is still acceptable. The generated figure is as follows. Note that the vector value has been scaled to the display.

In this way, the calculation requires the st distribution, that is, the object direction cannot be reversed, and P cannot be continuous in the St space. Therefore, there must be a piece of model that will not be continuous.Or more intelligently, bake has the same objectid, so that the tangent is calculated independently in each part of the object.