Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Anyone knows what's the meaning of odometry_benchmark and odometry_benchmark_img? #80

Open
robocar2018 opened this issue Feb 23, 2023 · 1 comment

Comments

@robocar2018
Copy link

Why we can only test the data which is included in the keys of **odometry_benchmark**. ? Like the below code shows

if dataset_name not in dataset.odometry_benchmark.keys():
            continue

# training set to the raw data of the KITTI dataset.
    # The following dict lists the name and end frame of each sequence that
    # has been used to extract the visual odometry / SLAM training set
    odometry_benchmark = OrderedDict()
    odometry_benchmark["2011_10_03_drive_0027_extract"] = [0, 45692]
    odometry_benchmark["2011_10_03_drive_0042_extract"] = [0, 12180]
    odometry_benchmark["2011_10_03_drive_0034_extract"] = [0, 47935]
    odometry_benchmark["2011_09_26_drive_0067_extract"] = [0, 8000]
    odometry_benchmark["2011_09_30_drive_0016_extract"] = [0, 2950]
    odometry_benchmark["2011_09_30_drive_0018_extract"] = [0, 28659]
    odometry_benchmark["2011_09_30_drive_0020_extract"] = [0, 11347]
    odometry_benchmark["2011_09_30_drive_0027_extract"] = [0, 11545]
    odometry_benchmark["2011_09_30_drive_0028_extract"] = [11231, 53650]
    odometry_benchmark["2011_09_30_drive_0033_extract"] = [0, 16589]
    odometry_benchmark["2011_09_30_drive_0034_extract"] = [0, 12744]
    
    #odometry_benchmark["2011_09_29_drive_0071_extract"] = [0, 8000]

    odometry_benchmark_img = OrderedDict()
    odometry_benchmark_img["2011_10_03_drive_0027_extract"] = [0, 45400]
    odometry_benchmark_img["2011_10_03_drive_0042_extract"] = [0, 11000]
    odometry_benchmark_img["2011_10_03_drive_0034_extract"] = [0, 46600]
    odometry_benchmark_img["2011_09_26_drive_0067_extract"] = [0, 8000]
    odometry_benchmark_img["2011_09_30_drive_0016_extract"] = [0, 2700]
    odometry_benchmark_img["2011_09_30_drive_0018_extract"] = [0, 27600]
    odometry_benchmark_img["2011_09_30_drive_0020_extract"] = [0, 11000]
    odometry_benchmark_img["2011_09_30_drive_0027_extract"] = [0, 11000]
    odometry_benchmark_img["2011_09_30_drive_0028_extract"] = [11000, 51700]
    odometry_benchmark_img["2011_09_30_drive_0033_extract"] = [0, 15900]
    odometry_benchmark_img["2011_09_30_drive_0034_extract"] = [0, 12000]

@scott81321
Copy link

scott81321 commented Apr 10, 2023

Hello @robocar2018

This is something better answered by Martin Brossard but I will give it a shot. Notice there is a discrepancy between the values of the noise covariances in the main routine, main_kitti.py, and those values mentioned in his paper (or thesis). The values in the main routine are generally much lower, especially the biases. E.g. cov_b_omega = 1e-8 in main_kitti.py but is 10^(-4) rad/s in the paper (that's 4 orders of magnitude lower).
What I suspect is that Brossard originally used the values of his paper and once he selected his trained data, reduced these values to get the best results. So these noise covariances in the program are set for his choice of training data. These noise covariances are likely too low for other kinds of data. From experience. I know these noise covariances have to be increased if you use less precise sensors than OXTS.

The program is not completely consistent I am afraid. E.g. there is (or was) a discrepancy in the format of stored CNN and the input format for reading it - a bug in effect (see issue entitled "Train filter failure with error: TypeError: zeros() received an invalid combination of arguments - got (NoneType, int, int), #72" and also issue "train parameters size mismatch #69"). You also have noise covariances in utils_numpy_filter.py but their values are very different from main_kitti.py and also they do not seem to matter so much (if at all?) . From experience, tests show me that it's the main_kitti.py that largely decides the initial Q matrix and P_0 matrix. Brossard got his program working well for his selection of test cases but it was done with some improvisations - likely an issue of time - ie the code is not really general purpose.

Mind you, I could use some feedback on the matter of noise covariances.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants