Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EaDeblur-GS: Event assisted 3D Deblur Reconstruction with Gaussian Splatting

Yuchen Weng, Zhengwen Shen, Ruofan Chen, Qi Wang, Jun Wang

TL;DR

The paper addresses severe motion blur in 3D reconstruction with NeRF/3DGS by fusing blurred RGB frames and event streams. It introduces EaDeblur-GS, which uses the Event Double Integral (EDI) to generate latent sharp views, COLMAP for pose refinement, and an Adaptive Deviation Estimator (ADE) to shift Gaussian centers via $\delta x_j^{(i)}$, supervised through Blurriness Loss and Event Integration Loss. Key contributions include the ADE module, loss designs, and demonstrated improvements over Deblur-GS and related baselines while preserving real-time rendering capability on synthetic data. The approach enhances robustness to motion blur in neural 3D representations, enabling high-fidelity interactive rendering in dynamic scenes.

Abstract

3D deblurring reconstruction techniques have recently seen significant advancements with the development of Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Although these techniques can recover relatively clear 3D reconstructions from blurry image inputs, they still face limitations in handling severe blurring and complex camera motion. To address these issues, we propose Event-assisted 3D Deblur Reconstruction with Gaussian Splatting (EaDeblur-GS), which integrates event camera data to enhance the robustness of 3DGS against motion blur. By employing an Adaptive Deviation Estimator (ADE) network to estimate Gaussian center deviations and using novel loss functions, EaDeblur-GS achieves sharp 3D reconstructions in real-time, demonstrating performance comparable to state-of-the-art methods.

EaDeblur-GS: Event assisted 3D Deblur Reconstruction with Gaussian Splatting

TL;DR

The paper addresses severe motion blur in 3D reconstruction with NeRF/3DGS by fusing blurred RGB frames and event streams. It introduces EaDeblur-GS, which uses the Event Double Integral (EDI) to generate latent sharp views, COLMAP for pose refinement, and an Adaptive Deviation Estimator (ADE) to shift Gaussian centers via , supervised through Blurriness Loss and Event Integration Loss. Key contributions include the ADE module, loss designs, and demonstrated improvements over Deblur-GS and related baselines while preserving real-time rendering capability on synthetic data. The approach enhances robustness to motion blur in neural 3D representations, enabling high-fidelity interactive rendering in dynamic scenes.

Abstract

3D deblurring reconstruction techniques have recently seen significant advancements with the development of Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Although these techniques can recover relatively clear 3D reconstructions from blurry image inputs, they still face limitations in handling severe blurring and complex camera motion. To address these issues, we propose Event-assisted 3D Deblur Reconstruction with Gaussian Splatting (EaDeblur-GS), which integrates event camera data to enhance the robustness of 3DGS against motion blur. By employing an Adaptive Deviation Estimator (ADE) network to estimate Gaussian center deviations and using novel loss functions, EaDeblur-GS achieves sharp 3D reconstructions in real-time, demonstrating performance comparable to state-of-the-art methods.
Paper Structure (8 sections, 8 equations, 2 figures, 3 tables)

This paper contains 8 sections, 8 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Adaptive Deviation Estimator
  4. Loss Functions
  5. Experiments
  6. Comparisons and Results
  7. Ablation studies
  8. Conclusion

Figures (2)

  • Figure 1: The Overview of our approach. Our approach integrates blurred RGB images with event data to enhance image clarity. The EDI technique is applied to produce latent sharp images, which are then refined through COLMAP for accurate SFM reconstruction and 3D Gaussian modeling. The ADE module estimates positional deviations based on initial Gaussian positions and camera extrinsics, simulating camera motion. These deviated Gaussians are rendered into multiple views and compute blurriness and event integration losses, facilitating the learning of detailed 3D representations for enhanced reconstruction quality.
  • Figure 2: Qualitative results on E2NeRF synthetic dataset.