A question about equation (23) in the paper

[YuePanEdward]

In the paper, the absolute trajectory error can be calculated as Equation 23.

Note that in the equation for the position error, there's the rotation error multiplying with the aligned position.

However, according to the code, the position error is calculated as:

e_trans_vec = (p_gt-p_es_aligned)
e_trans = np.sqrt(np.sum(e_trans_vec**2, 1))

without the multiplication of the rotation error.

I wonder which one is correct.

[BambValley]

Just personal opinion.

I think the version without the multiplication of the rotation error is correct.

In the paper , the symbol R and p is used in Equation (2) in such way:

It can be deduced that R_i denotes the rotation from camera to world and p_i denotes the position of system in the world at time i.

Then, the error between p_i and p_i' is naturally Euclidean distance, since they are in the same coordinate.