lanwu076 / log-gpis-demo Goto Github PK

Hello there!
Even though the code runs smoothly and I am able to recreate a few figures from your paper, I have not completely understood some parts. Could you clarify a few things, please?

1. Covariance functions cov1 and cov2 are somewhat different from the equations presented in the paper. Specially cov2. From my understanding, with scale = sqrt(2*v) from line 29, you get that (sqrt(2*v))*(lambda/scale) = lambda. From this I have two questions:
   1.1. Why is lambda in the numerator?
   1.2. Why is it done this way? Have I missed any point here?

2. In lines 63-64, when creating the data vector:
   2.1. Why do you subtract 0.05?
   2.2. Why is y = exp(-y*lambda)? I couldn't find anything related to this in the paper.

Thank you for the attention!

In the paper you write

While the resulting EDF now loses the sign, it is possible to recover the sign using the surface normal and sensor’s location.

and

However, we recover the sign of the Log-GPIS by comparing the predicted gradient of each testing point with sensor position. If the gradient is in the opposite direction of the sensor, we flip the sign of distance value of the testing point.

I could not find throughout the paper how you define "sensor position". Looking at the source code has also led me nowhere. Could you give a hand, please? How would I reproduce it in the source code you provide?