JADE-GS: Joint Alternating Deblurring Guided by Events in 3D Gaussian Splatting

2026-07-16

AI Summary

This work addresses the challenge of image blur induced by fast camera motion, which degrades the sharp visual cues essential for accurate 3D reconstruction. While event cameras offer high temporal resolution, they struggle to provide reliable 3D supervision signals. To overcome these limitations, the authors propose a pixel-adaptive gating fusion mechanism that integrates physics-driven and learning-based deblurring priors. They further introduce a bidirectional recurrent optimization framework coupling a 2D deblurring module with a 3D Gaussian splatting model, jointly refining geometry and texture through multi-view consistency rendering and physical re-blurring constraints. This approach mitigates error propagation and pseudo-label bias inherent in conventional unidirectional pipelines. Experiments demonstrate state-of-the-art perceptual quality on both synthetic and real-world datasets—achieving leading scores in LPIPS and CLIP-IQA while maintaining competitive PSNR and SSIM—and enable training within approximately one hour on a single consumer-grade GPU with real-time rendering capability.

Abstract

When a camera moves fast during exposure, blur destroys the intra-exposure motion a 3D model needs to recover the sharp scene, while event cameras capture exactly this signal at microsecond resolution. Turning them into reliable 3D supervision faces two obstacles. First, the two restoration priors fail in opposite ways: physics-based event-integration priors preserve edges but accumulate drift; learned networks recover texture but distort boundaries. Second, existing pipelines run in one direction only, so raw event noise or the biases of fixed 2D pseudo-labels pass uncorrected into the geometry. JADE-GS addresses both: a pixel-adaptive routing gate fuses the complementary priors, and the resulting 2D restorer is coupled to a 3D Gaussian Splatting student in a bidirectional loop, where detached, multi-view-consistent renders and a physics-based reblurring constraint regularize the restorer, turning a fixed preprocessor into a geometry-aware predictor. Across synthetic and real benchmarks, JADE-GS attains the best perceptual quality, leading LPIPS and CLIP-IQA on both benchmarks with competitive PSNR and SSIM, and trainsin about one hour under 5 GB on a single consumer GPU while preserving real-time rendering.

Problem

• event cameras
• motion deblurring
• 3D Gaussian Splatting
• 3D reconstruction
• image restoration

Innovation

• event camera
• 3D Gaussian Splatting
• joint deblurring
• bidirectional optimization
• pixel-adaptive fusion