rosbag

[dongfangzhou1108]

appreciate your great work！
can you release your rosbag in paper？
and have you test，can your method work in the condition of white wall？
thanks a lot！

[lastflowers]

Hi, thank you for your interest in our work!

Currently, I am not sure to release the flight dataset in the paper due to a license issue.
I expect that localizing in a pure white wall would be very challenging even for our method since our visual front-end depends on image gradients.

If you are interested, please check our snu301 sequences with a textureless wall in the readme file.

[dongfangzhou1108]

thanks a lot.
but when i try your snu301 rosbag, i find some wrong in your right camera like picture below.
best wishes!

[arenas7307979]

There seems to be an error about "cam1 distortion_coeffs ", I tried to fill in the same parameter as cam0 and it works fine, the correct value can wait for to update from author

[lastflowers]

I found that this is due to a different implementation of cv::fisheye::estimateNewCameraMatrixForUndistortRectify in opencv versions.

To resolve the issue, you should use the above function in OpenCV >= 3.4
https://github.com/opencv/opencv/blob/3.4/modules/calib3d/src/fisheye.cpp#L549

The new intrinsic matrix should be

cv_Kl_new = [300.3127324163617, 0, 346.0204157455623;
0, 299.9143972488393, 229.9477505201118;
0, 0, 1]

cv_Kr_new = [300.290757544907, 0, 379.7844651209092;
0, 300.1389715440906, 242.1087340715927;
0, 0, 1]

Or you can just put this value to run the custom dataset.