How can I run main_slam.py to generate estimated trajectory file?

[luigifreda]

How can I run main_slam.py to generate estimated trajectory file?

[linpeisensh]

It's weird for me.
The ATE of slam is bigger than vo.

Here is my definition of trajectory of slam:

trajectory = []
n_his = len(st_his.relative_frame_poses)
for i in range(n_his):
	if st_his.slam_states[i] == SlamState.OK: #OK
		cur_pose = CameraPose(st_his.relative_frame_poses[i])
		cur_tra = [str(round(st_his.timestamps[i], 4))] + list(map(str, np.round(cur_pose.Ow, decimals=4))) + \
				  list(map(str, np.round(R.from_matrix(cur_pose.Rcw).as_quat(), decimals=4)))
		trajectory.append(cur_tra)

Do I misunderstand something?
@luigifreda
Thank you in advance!

[luigifreda]

Hi,

1. you cannot compare main_slam.py with main_vo.py.
   Did you study the two methods? As explained in the README

main_vo.py combines the simplest VO ingredients without performing any image point triangulation or windowed bundle adjustment. At each step $k$, main_vo.py estimates the current camera pose $C_k$ with respect to the previous one $C_{k-1}$. The inter-frame pose estimation returns $[R_{k-1,k},t_{k-1,k}]$ with $||t_{k-1,k}||=1$. With this very basic approach, you need to use a **ground truth** in order to recover a correct inter-frame scale $s$ and estimate a valid trajectory by composing $C_k = C_{k-1} * [R_{k-1,k}, s t_{k-1,k}]$. This script is a first start to understand the basics of inter-frame feature tracking and camera pose estimation.

Therefore, main_vo.py is using the ground truth in order to retrieve the interframe scale!
On the other hand, main_slam.py is estimating the trajectory up-to-scale given it's a pure monocular approach, without using any information coming from the ground truth.

2. you cannot compare the slam/vo trajectories directly with the ground truth.
   In order to compare trajectories, you need to align them by using a package like
   https://github.com/MichaelGrupp/evo
   and considering the manifolds ( SE(3) or Sim(3) ) in which the methods compute their estimates.

3. in slam.tracking.tracking_history there are keyframes references and relative frame poses.
   There is not a function that dumps the final computed trajectory at the end of the run (after all the bundle adjustments). But it easy to code it if you need it.

[linpeisensh]

Thank you for your reply!
Here is my estimation trajectory code after slam process is done :

trajectory = []
n_his = len(st_his.relative_frame_poses)
for i in range(n_his):
	if st_his.slam_states[i] == SlamState.OK: #OK
		cur_pose = CameraPose(st_his.relative_frame_poses[i])
		cur_tra = [str(round(st_his.timestamps[i], 4))] + list(map(str, np.round(cur_pose.Ow, decimals=4))) + \
				  list(map(str, np.round(R.from_matrix(cur_pose.Rcw).as_quat(), decimals=4)))
		trajectory.append(cur_tra)

Could you help me to check, whether it is correct? Or could you show me what could you generate this trajectory file?