load input image and convert it to a (3, H, W) tensor with RGB values in [0, 1]

Abstract

AnyCalib: On-Manifold Learning for Model-Agnostic Single-View Camera Calibration (ICCV 2025)