Supplementary Material for E³NeRF: Efficient Event-Enhanced Neural Radiance Fields from Blurry Images

Area of Science:

• Computer Vision
• 3D Reconstruction
• Neural Rendering

Background:

• Neural Radiance Fields (NeRF) excel at novel view synthesis but struggle with blurry input data common in real-world scenarios.
• Existing methods often fail to produce sharp NeRF reconstructions from motion-blurred or low-light images.

Purpose of the Study:

• To develop an Efficient Event-Enhanced NeRF (E³NeRF) framework for reconstructing sharp NeRF models from blurry images.
• To leverage event streams alongside blurry images to overcome limitations in traditional NeRF reconstruction.
• To improve the efficiency and applicability of NeRF in challenging real-world conditions.

Main Methods:

• Proposed E³NeRF framework integrating blurry images and event streams for NeRF training.
• Introduced blur rendering loss and event rendering loss to model physical processes.
• Utilized latent spatial-temporal blur information from event streams for efficient training.
• Developed a camera pose estimation framework guided by events for real-world data.

Main Results:

• E³NeRF successfully reconstructs sharp NeRF models from blurry input images.
• The framework demonstrates superior performance in scenes with high-speed non-uniform motion and low-light conditions.
• Experiments on synthetic and real-world data validate the effectiveness of the proposed approach.
• Achieved more profound utilization of the relationship between events and images compared to prior works.

Conclusions:

• E³NeRF offers a robust solution for generating high-quality NeRF from challenging blurry image data.
• The integration of event streams significantly enhances NeRF reconstruction accuracy and efficiency.
• The method generalizes well to practical applications, particularly in dynamic and low-illumination environments.