cell's center, covariance and feature_direction

[narutojxl]

Hi @ziv-lin , thanks for sharing your work to community. when calculating cell's center cell_map_keyframe.hpp func find_cell_center() and covariance func get_covmat() in code is as follows, different from the paper equ1. and equ3. repectively, does this matter?

cell_center(0) = (std::round((pt(0) - half_of_box_size) / box_size)) box_size + half_of_box_size; cell_center(1) = (std::round((pt(1) - half_of_box_size) / box_size)) box_size + half_of_box_size; cell_center(2) = (std::round((pt(2) - half_of_box_size) / box_size)) * box_size + half_of_box_size;

for (size_t i = 0; i < pt_size; i++) { m_cov_mat = m_cov_mat + (m_points_vec[i] m_points_vec[i].transpose()).template cast(); } m_cov_mat -= pt_size (m_mean * m_mean.transpose()); m_cov_mat /= (pt_size - 1);

When calculating feature direction in func feature_direction(), in paper V-C(2D histogram of keyframe), " we choose the direction with positive X components", in code we transform points whose x part is negative into positive, could you please explain it? Looking forward to your reply, thanks for your help!

if (vec_3d[0] < 0) { vec_3d *= (-1.0); }

[ziv-lin]

Since the direction in 3D space have two directions, for convenience, we choose the direction with its x value alway positive. In addition, we have mentioned this point in our preprint paper:)

"we choose the direction with positive X components, i.e., Cd = sign(C{d_x}) · C_d."

[ziv-lin]

The compute of covariance is correct, which is more effective that can save more time in computation.

[narutojxl]

Hi @ziv-lin , thanks for your reply! For covariance calculate, i don't understand why paper equ3. equal to the format in code, could you please give some tips? For calculate histogram of keyframe, you say "we choose the direction with positive X components, i.e., Cd = sign(C{d_x}) · C_d.". I understand what you mean is for negative x component, we should ignore these cells, we only use positive x of cells to calculate histogram. Did I get it wrong? For center i see the code also slightly different from the paper equ1.

[narutojxl]

There is a point p(black) in the bound of the following picture, which constructed with the center point(red) and block size. I found that some black points doesn't belong to the cell, because they created cell's center does not equal to the red one. Is that right? The correspongding code is find_cell().