What is the reason for calculating covariance with mean[3] = height in Kalman filter?

[YoungjaeDev]

ByteTrack/yolox/tracker/kalman_filter.py

Lines 107 to 116 in 8d47ea5

	std_pos = [
	self._std_weight_position * mean[3],
	self._std_weight_position * mean[3],
	1e-2,
	self._std_weight_position * mean[3]]
	std_vel = [
	self._std_weight_velocity * mean[3],
	self._std_weight_velocity * mean[3],
	1e-5,
	self._std_weight_velocity * mean[3]]

Why is there a special reason why I took it as a height when modeling?

[ydm365]

As indicated by authors' comment, the position array is coded by order x, y, a, h. The forth number(mean[3]) is scale of the bounding box, and I think they assume larger object makes greater movement, and greater dilation or contraction. Actually the other three numbers is independent with scale of objects.

[YoungjaeDev]

@ydm365
In other words, a cov is given to pos that changes in real time in proportion to the object size

[ydm365]

@ydm365 In other words, a cov is given to pos that changes in real time in proportion to the object size

As far as I can tell, this is exactly what the authors are assuming.