Unstable Estimates when running Intrinsic & Stereo Calib

[ank95]

Hi
I am trying to do Intrinsic and Stereo Calibration of fisheye lens using the Mei Rives model but there seems to be a huge deviation in the mirror parameter(xi) and gamma values.
Using different data sets, xi deviates anywhere from 1.2 to 2.1 and gamma values deviate between 1200 to approximately 1600.

There also seems to be a difference in the intrinsic values estimated by Intrinsic Calib and Stereo Calib for the same dataset. The extrinsics estimated by the Stereo Calib are quite comparable through-out different datasets.

Final Reprojection Error is mostly close to 0.3.

Any help would be really appreciated.

[poaongithub]

@ank95, same, have you solved it? thanks in advance

[abhilash-nandkumar]

I ended up using the Kalibr package from ethz-asl. Their omnidirectional model produced more satisfactory results.