PRISM3D: Probabilistic Refinement and Robust Initialization
for Physically Consistent Scene Modeling
under Extreme Motion Blur

ECCV 2026

Gopi Raju Matta1 Trisha Reddypalli1 Divya Madhuri Vemunuri1 Kaushik Mitra1
1Indian Institute of Technology Madras
Paper (PDF) Code Coming Soon Dataset Coming Soon
Input PRISM3D PRISM3D-E

PRISM3D framework learns a sharp 3D Gaussian representation of the scene along with its camera motion trajectories directly from extreme motion-blurred images, enabling state-of-the-art deblurring, high-quality novel view synthesis, and real-time rendering, all with efficient training time and minimal memory overhead.

Why PRISM3D?

Directly Handles Blur

Unlike previous approaches, PRISM3D reconstructs scenes without requiring sharp reference images.

Robust Initialization

VGGSfM and MCMC-based Gaussian initialization enable stable optimization under extreme blur.

Event-Based Extension

PRISM3D-E leverages event cameras for improved reconstruction when event streams are available.


Abstract

We address the inverse problem of blind 3D scene reconstruction from extremely motion-blurred images, a scenario where traditional Structure-from-Motion pipelines fail. Unlike traditional methods that require sharp images for pose estimation, PRISM3D directly processes blurred images, making it practical for real-world scenarios.

Our approach utilizes a Robust Initialization strategy via deep dense tracking (VGGSfM) to recover global topology where feature matching fails. To robustly populate these sparse priors, we adopt a probabilistic formulation for geometric densification via Markov Chain Monte Carlo (MCMC), while simultaneously modeling physical image formation via continuous Bézier Trajectories.

Furthermore, we introduce PRISM3D-E, a multi-modal (RGB + Events) extension that seamlessly integrates high-temporal-resolution events (EDI deblurring) as structural priors to maximize geometric recovery. To facilitate future research, we concurrently contribute the PRISM3D-E Benchmark dataset specifically curated for extreme blur scenarios.


PRISM3D Pipeline


PRISM3D Pipeline Overview

Overview of PRISM3D and PRISM3D-E. Our framework jointly estimates camera motion and a sharp Gaussian representation directly from severely motion-blurred images, utilizing deep SfM for robust initialization and MCMC for probabilistic geometric refinement.


PRISM3D Qualitative Results


PRISM3D and its event-based extension PRISM3D-E achieve high-quality 3D reconstructions and sharp novel view synthesis under extreme motion blur. Our results on both synthetic and real-world datasets demonstrate state-of-the-art performance in challenging blur scenarios.

Real-World Dataset Results

Input PRISM3D PRISM3D-E
Camera
Lego
Letter
Plant
Toys


Synthetic Dataset Results

Hover over the images below to view the outputs of our method!

PRISM3D (Standalone)



PRISM3D-E (Event-Assisted)

PRISM3D Comparisons


We compare PRISM3D and PRISM3D-E against strong baselines including MPRNet, Restormer, EDI, E2NeRF, and EBAD-NeRF on both synthetic and real datasets. Our method produces sharper reconstructions with better detail preservation under extreme motion blur. Note that ExBluRF* and BAD-Gaussians* rely on sharp-image-based initialization and are excluded from real-world comparisons due to their impracticality.


Real-World Dataset

Real-world comparisons


Synthetic Dataset

Synthetic comparisons


BibTeX

@inproceedings{matta2026prism3d,
      author       = {Matta, Gopi Raju and Reddypalli, Trisha and Vemunuri, Divya Madhuri and Mitra, Kaushik},
      title        = {{PRISM3D: Probabilistic Refinement and Robust Initialization for Physically Consistent Scene Modeling under Extreme Motion Blur}},
      booktitle    = {European Conference on Computer Vision (ECCV)},
      year         = {2026}
    }