Question: Why Convert Intrinsics Between OpenCV and COLMAP in `camera_matrix_of_crop`?

I was looking at the camera_matrix_of_crop function and noticed that it converts the camera intrinsics from OpenCV to COLMAP format, does some scaling and translation, and then converts it back to OpenCV. I'm wondering why this conversion is needed. Is there a reason we can't just do these operations directly in OpenCV format? Just trying to figure out if this step is really necessary or if it might be a bit redundant.

Thanks for any insights!

https://github.com/naver/dust3r/blob/c9e9336a6ba7c1f1873f9295852cea6dffaf770d/dust3r/datasets/utils/cropping.py#L95-L98