I have some question about 'Self-Supervised Camera Self-Calibration from Videos'.
Using code in this repo., I add implementation of UCM to monodepth2 code and test on kitti dataset to confirm that UCM can converge to kitti pinhole model. But the result can not converge to GT pinhole camera parameter.
I your paper, It can be converge GT parameter quickly even though with parameter initialization using image shape.
In your test, how many epochs need to converge to around GT camera parameter?
Hi, thanks for your great research!!
I have some question about 'Self-Supervised Camera Self-Calibration from Videos'. Using code in this repo., I add implementation of UCM to monodepth2 code and test on kitti dataset to confirm that UCM can converge to kitti pinhole model. But the result can not converge to GT pinhole camera parameter.
I your paper, It can be converge GT parameter quickly even though with parameter initialization using image shape. In your test, how many epochs need to converge to around GT camera parameter?