question about scalability #5
Description
In your paper you said that "regressing the relative pose does not require scene-specific training of the network, which improves scalability, and it can also be applied to scenes which are not available during the training of the network".
Indeed the relative pose is not dependent on the training scene. Howerer, given an image pair, we can get the relative pose according to Epipolar Geometry and it is relative with the camera intrinsics.
Is the model able to apply to test scene when using different camera in test scene?