DiffusionBAS: Estimating Camera External Orientation Through Diffusion Process

Abstract. Violent forces such as earthquakes or human interaction can damage or demolish objects of cultural heritage. Many architectural masterpieces have survived only in a few photos or drawings. Moreover, often interior decorations of buildings such as stucco or paintings are destroyed by fire. Therefore, an automatic 2D-to-3D reconstruction that can assists architecture historians in process of restoration of original 3D shape of a lost site of culture heritage is required. An automatic in-paint method can assists restoration of partially destroyed stucco and paintings. We present an end-to-end framework that receives a single image of an object and predicts its vector of the exterior orientation parameters. The main objective of a present work is reconstruction of 3D model of a partially destroyed 3D object and it 3D in-painting. As the initial and prerequisite phase of this framework we propose a new generative model for estimation of the exterior orientation parameters for a given input image using a Diffusion model (DiffusionBAS).