Open rakhimovv opened 4 months ago
Hi! Thank you for your work!
In the equation (9) in the paper:
\tilde{\alpha}(z_0,z_1)=\frac{t(z_1)-t(z_0)}{1-t(z_0)}
the denominator = 1-t(z_0)
1-t(z_0)
While in code: alpha_value = torch.log(hit_prob / (visibility - hit_prob + eps) + eps) If we omit log the denominator = visibility - hit_prob = 1 - t(z_0) - (t(z_1) - t(z_0)) = 1-t(z_1)
alpha_value = torch.log(hit_prob / (visibility - hit_prob + eps) + eps)
denominator = visibility - hit_prob = 1 - t(z_0) - (t(z_1) - t(z_0)) = 1-t(z_1)
The denominator in the paper and the code do not match.
Is this intended behaviour? and where is the error in the paper or the code or none?
Hi, this is an estimation of the alpha value. In an ideal continuous case, we need to divide the 1-z at z(t). But in discrete cases, we may divide by either the left point or right point here.
Hi! Thank you for your work!
In the equation (9) in the paper:
\tilde{\alpha}(z_0,z_1)=\frac{t(z_1)-t(z_0)}{1-t(z_0)}
the denominator =
1-t(z_0)
While in code:
alpha_value = torch.log(hit_prob / (visibility - hit_prob + eps) + eps)
If we omit log thedenominator = visibility - hit_prob = 1 - t(z_0) - (t(z_1) - t(z_0)) = 1-t(z_1)
The denominator in the paper and the code do not match.
Is this intended behaviour? and where is the error in the paper or the code or none?