Perspective camera model rays?

[paulmelis]

Not sure this is a bug, but I noticed the visualization of camera rays (pixel centers?) on https://docs.nerf.studio/en/latest/nerfology/model_components/visualize_cameras.html:

To me this doesn't make much sense, with each ray (or pixel center) apparently having a fixed angular offset compared to its neighbors. Shouldn't this be a perspective projection that assumes a flat image plane in front of the camera, with a uniform pixel width, causing varying angles between the rays?

The current visualization for the perspective camera is pretty much the same as for the fisheye model, which should not be the case. But maybe I'm misunderstanding what that visualization is supposed to show and/or you're doing some form of weighting per ray based on the pixel it contributes to. Also, I did notice the CamerasV2 branch, so maybe this is something that's being worked on.