units of distortion parameters in the brown conrady camera model

I want to replace the image metadata extraction step with my own meta data - this seems straightforward, as I can get all these (focal_x, focal_y, c_x, c_y) in pixels (from my own metadata extraction) and then convert to scaled coordinates by dividing by max(width, height).

I do have an initial set of distortion parameters (k1,k2,k3, p1,p2) for the brown conrady model, but these are in pixel coordinates - when scaling these as well, I get a lot of non-convergence errors from Ceres in the reconstruction step - naturally, also these parameters would also be quite large in scaled coordinates ( e.g. k3 -> k3*width^6), but I haven't seen any such large parameter values posted here - Can you clarify in what units openSFM expects these parameters ?

Start by tracing the image metadata extraction and reconstruction steps described in the issue, including the focal, principal-point, and Brown-Conrady distortion parameters. Confirm the units expected by the camera model and whether scaling the distortion values causes the reported Ceres non-convergence; document the verified units and conversion guidance.