Neural Implicit Surface Reconstruction from Noisy Camera Observations

TL;DR

This paper introduces a method for reconstructing 3D surfaces from noisy camera data by jointly learning camera parameters and surface representations, enabling accurate 3D modeling without precise calibration.

Contribution

It presents a novel approach that jointly learns camera parameters and surface representations, reducing the need for accurate camera calibration in 3D reconstruction.

Findings

Effective reconstruction with noisy camera data

Joint learning improves surface quality

Robustness to calibration errors

Abstract

Representing 3D objects and scenes with neural radiance fields has become very popular over the last years. Recently, surface-based representations have been proposed, that allow to reconstruct 3D objects from simple photographs. However, most current techniques require an accurate camera calibration, i.e. camera parameters corresponding to each image, which is often a difficult task to do in real-life situations. To this end, we propose a method for learning 3D surfaces from noisy camera parameters. We show that we can learn camera parameters together with learning the surface representation, and demonstrate good quality 3D surface reconstruction even with noisy camera observations.