Test doesn't need the trained model?

[yzhang93]

Hello,

I found your paper very interesting and thanks for sharing the codes! I test the filters by running main_kitti.py, and got similar results as in paper. However, I didn't find in codes where you load the trained model for test. I deleted the provided training parameters (delta_p.p, iekfnets.p, normalize_factors.p), and it still generates the same results. Did you hard coded your trained parameters somewhere?

[mbrossar]

Hi,

Thanks you for you interrest.
I did not hard coded trained parameters somewhere, they are included in iekfnets.p.

delta_p.p and normalize_factors.p are two files that avoid useless computation. If you delete it, the code will just compute them from data.

If you delete iekfnets.p, you will have a filter without training parameters. The filter still works well but it is better once trained (See Section V-D in the paper). This is particularly remarkable.

[yzhang93]

@mbrossar Thanks for your reply!

I'm also wondering how did you set the parameters in class KITTIParameters?
cov_omega = 2e-4 cov_acc = 1e-3 cov_b_omega = 1e-8 cov_b_acc = 1e-6 cov_Rot_c_i = 1e-8 cov_t_c_i = 1e-8 cov_Rot0 = 1e-6 cov_v0 = 1e-1 cov_b_omega0 = 1e-8 cov_b_acc0 = 1e-3 cov_Rot_c_i0 = 1e-5 cov_t_c_i0 = 1e-2 cov_lat = 1 cov_up = 10
And these values are different compared to those in class Parameters in utils_numpy_filter.py, and also different from the initial values stated in the paper.
`# Process noise covariance
cov_omega = 1e-3
"""gyro covariance"""
cov_acc = 1e-2
"""accelerometer covariance"""
cov_b_omega = 6e-9
"""gyro bias covariance"""
cov_b_acc = 2e-4
"""accelerometer bias covariance"""
cov_Rot_c_i = 1e-9
"""car to IMU orientation covariance"""
cov_t_c_i = 1e-9
"""car to IMU translation covariance"""

    cov_lat = 0.2
    """Zero lateral velocity covariance"""
    cov_up = 300
    """Zero lateral velocity covariance"""

    cov_Rot0 = 1e-3
    """initial pitch and roll covariance"""
    cov_b_omega0 = 6e-3
    """initial gyro bias covariance"""
    cov_b_acc0 = 4e-3
    """initial accelerometer bias covariance"""
    cov_v0 = 1e-1
    """initial velocity covariance"""
    cov_Rot_c_i0 = 1e-6
    """initial car to IMU pitch and roll covariance"""
    cov_t_c_i0 = 5e-3
    """initial car to IMU translation covariance"""

`

[Xiaojuan6893]

Hello,

I found your paper very interesting and thanks for sharing the codes! I test the filters by running main_kitti.py, and got similar results as in paper. However, I didn't find in codes where you load the trained model for test. I deleted the provided training parameters (delta_p.p, iekfnets.p, normalize_factors.p), and it still generates the same results. Did you hard coded your trained parameters somewhere?

Hello, Mr/Ms. Zhang!

Sorry to interrupt, I have tried to download KITTI IMU raw data many times, but I have been unable to succeed. Would you mind sharing it like Baidu Cloud?
Thank you!

[mbrossar]

Hello,

You can find my reformatted kitti data in google drive : https://drive.google.com/open?id=1DClhQEDayv2p4IJ1a9XydF-FpnWh5_EQ

You can also find if required temp.zip at https://drive.google.com/open?id=1WPuC71kSb-dq0gSjrrSwWt8WM7XepKmd

Martin
Martin