Camera Calibration - Different image sizes

To apply solvePnP I have to do a camera calibration. I have received calibration images of size 4032x3024, while the images where I have to apply solvePnP are of size 2348x3024. They were made by the same camera (iPhone SE, 2nd edition). Will the resulting calibration data be applicable to the smaller image, on which I have to apply solvepnp?

Sidenote: The calibration images taken by pointing the camera to a screen, showing this image. Is that already a problem, since it was not printed on a A4 paper.

how were the smaller images made from complete sensor data? I have an idea but I want you to know this.

using a calibration pattern displayed on a screen is a good idea because screens are made to better tolerances than a random printer can print. you should make sure the pattern is scaled to match the display’s native resolution, so the squares sit precisely on pixel boundaries. otherwise you’ll have inaccuracies on the order of the size of a pixel.

crosspost: opencv - Camera Calibration - different image sizes - Stack Overflow

I currently suspect that there was cropping involved in the smaller image. Maybe I could embed the smaller image in a larger empy one.

A few things come to mind:

  1. If it is a crop, you can embed your image in a larger empty one, assuming you know how the crop was done.
  2. It might be inconvenient / inefficient to process your images this way, so you could transform the camera matrix to match your cropped images. (Consider how the Cx Cy values will be different for different size images)
  3. I suggest you make sure it’s just a crop and there isn’t scaling taking place too.
  4. This calibration is only valid for the focus and zoom level that the calibration images were captured at. If focus it zoom changes, your calibration accuracy could suffer, possibly quite a lot.

if it’s a crop, and I suspect that too, all the intrinsics stay the same. well, focal length and distortion coefficients for sure… cx, cy will be shifted according to the crop parameters. and maybe a rotation by 90 degrees.

I compared the meta data of the original images and found this


I guess that last bit might be the nail in the coffin. My solvePnP results are currently also slightly off.

those values can be indications but don’t rely on them. verify them. those focal length “equivalents” are rounded and generally useless. those “zoom ratio” things also don’t strike me as trustworthy.

mount the phone in a vise or something, take the same picture in both modes, check the pictures.