EPro-PnP provides a continuous counterpart to the categorical Softmax layer for 6DoF pose estimation networks. It formulates perspective-n-points as an end-to-end learnable layer using Gauss-Newton and Levenberg-Marquardt optimization within the network, enabling gradient backpropagation through the geometric solver. The method achieved Best Student Paper at CVPR 2022 and is designed for monocular object pose estimation tasks.