Question about Density estimation algorithm about sky-optimization

[misaka-10032]

No. This is p_c itself. The intuition is pick a value somewhere below 1, because the density of a uniform distribution is 1. Choosing some value below one is just to encourage triggering, because we do want to see the sky in-between trees.

[misaka-10032]

To clarify, what we get here is a probability density, not probability itself, so it's not limited within [0, 1], as long as the integral is 1.
https://en.wikipedia.org/wiki/Probability_density_function

The range seems about right. The computation is done in the log scale, so the log range is [-inf..11].

Please be aware that the precision could be an issue if it's not handled properly for LogSumExp
https://en.wikipedia.org/wiki/LogSumExp

[andriyrizhiy]

So, p_i = log(p_i)? And for threshold pc = 0.6 in Section 3.2 you use log(p)?

[andriyrizhiy]

Understand. Thank you!