Learning to Remove Lens Flare in Event Camera

2025-12-09

AI Summary

Event cameras offer high temporal resolution and dynamic range but are highly susceptible to lens flare, introducing complex spatiotemporal artifacts into event streams—a long-overlooked challenge. To address this, we propose the first physics-driven forward model of event-domain lens flare and introduce E-Flare-2.7K (synthetic) and E-Flare-R (real-world), the first paired simulation–real benchmark for event flare mitigation. We further design E-DeflareNet, a lightweight convolutional-recurrent hybrid network for robust flare removal. Our contributions are threefold: (1) the first optical-principles-based synthetic training set and real-world paired test set for event flare; (2) a theoretical framework characterizing the nonlinear suppression behavior of event streams under flare corruption; and (3) state-of-the-art performance on both synthetic and real data, yielding significant improvements in downstream tasks—including optical flow estimation and object tracking. All code and datasets are publicly released.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Learning on the Edge & Model Compression

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Event cameras have the potential to revolutionize vision systems with their high temporal resolution and dynamic range, yet they remain susceptible to lens flare, a fundamental optical artifact that causes severe degradation. In event streams, this optical artifact forms a complex, spatio-temporal distortion that has been largely overlooked. We present E-Deflare, the first systematic framework for removing lens flare from event camera data. We first establish the theoretical foundation by deriving a physics-grounded forward model of the non-linear suppression mechanism. This insight enables the creation of the E-Deflare Benchmark, a comprehensive resource featuring a large-scale simulated training set, E-Flare-2.7K, and the first-ever paired real-world test set, E-Flare-R, captured by our novel optical system. Empowered by this benchmark, we design E-DeflareNet, which achieves state-of-the-art restoration performance. Extensive experiments validate our approach and demonstrate clear benefits for downstream tasks. Code and datasets are publicly available.

Problem

• Removing lens flare from event camera data
• Addressing spatio-temporal distortion in event streams
• Developing a physics-based framework for flare removal

Innovation

• Physics-based forward model for flare removal
• First paired real-world dataset E-Flare-R
• E-DeflareNet achieves state-of-the-art restoration