LSE-NeRF: Learning Sensor Modeling Errors for Deblured Neural Radiance Fields with RGB-Event Stereo

TL;DR

This paper introduces LSE-NeRF, a method that combines RGB images and event camera data to reconstruct clear Neural Radiance Fields despite fast camera motions and blur, by modeling sensor errors and learning mappings between modalities.

Contribution

The paper proposes a novel approach that models sensor imperfections and learns cross-modal mappings, enabling high-quality NeRF reconstructions with stereo RGB and event camera data.

Findings

Improved NeRF reconstructions with fast camera motion.

Introduction of a new stereo RGB-event dataset.

Enhanced performance over existing methods like EVIMOv2.

Abstract

We present a method for reconstructing a clear Neural Radiance Field (NeRF) even with fast camera motions. To address blur artifacts, we leverage both (blurry) RGB images and event camera data captured in a binocular configuration. Importantly, when reconstructing our clear NeRF, we consider the camera modeling imperfections that arise from the simple pinhole camera model as learned embeddings for each camera measurement, and further learn a mapper that connects event camera measurements with RGB data. As no previous dataset exists for our binocular setting, we introduce an event camera dataset with captures from a 3D-printed stereo configuration between RGB and event cameras. Empirically, we evaluate our introduced dataset and EVIMOv2 and show that our method leads to improved reconstructions. Our code and dataset are available at https://github.com/ubc-vision/LSENeRF.