RS calibration failing about kalibr HOT 2 CLOSED

Mmh, it is odd that you would get a negative line delay.
I was not fully sure about this statement of yours:

With a framerate of 20 it completes successfully, while if I try 6 if fails (Spline Coefficient Buffer Exceeded).

Do you mean when setting the script option? Luc's documentation suggests that this parameter provides some control over the knot sequence governing the spline: https://github.com/ethz-asl/kalibr/blob/master/aslam_offline_calibration/kalibr/python/kalibr_rs_camera_calibration/RsCalibrator.py#L70-L73
Accordingly, it might also somewhat depend on the motion and rather than just setting the exact framerate it might be worth varying the parameter a little.

Do you detect the target in every single frame? I am just guessing, but your problems could also arise from missing detections and consequently gaps in the support of the spline. In that case, you could also try the dot grid pattern as recommended in the respective paper.

Finally, calibrating the line delay without an IMU appears to be a much more difficult problem than when an IMU is present (at least to me). The reason for that lies in the fact that you would like to "constrain" the spline in a way that discourages the estimation of infeasible motions (think about it, if infinite linear and angular accelerations were a possibility, the camera could basically teleport itself to poses that minimize the reprojection error).
Luc addressed this concern with regularizer terms and some clever knot placement. However, if you had an IMU, you could directly "constrain" the accelerations. Accordingly, it might still be a valid approach to push ahead with the transfer of RS error terms into the camera/IMU calibration although the camera calibration currently fails for you.

from kalibr.

Yeah I meant "passing 20 as a framerate to the script".

Your suggestion about the features not always being detected was on point: I recorded a dataset making sure the full target was always visible and it started to run. I will have to do another one with more light, but it seems this was the main blocker. I guess the optimization was indeed trying to teleport ;)

I will try another dataset and then set to do the change to the code for cam + IMU.

Thanks again, very helpful.

from kalibr.