using different cameras

[bobluoluo]

the paper says that images are using the same camera parameters, sharing the same camera calibrations. Does it still work if I remove this constrain? Would it still work if one removes the assumption that images are sharing the same camera parameters?

[ziruiw-dev]

Hi @bobluoluo,

We didn't try to remove this constraint. You can try it easily by modifying the focal length from 2 learnable parameters to Nx2 learnable parameters (N is the number of images). I am curious to know the result as well. You are more than welcome to put the results here if you got them 🙂

I guess it should be able to handle certain FOV variances but might fail when the FOVs differences are too large.

Best,
Zirui