EeveeDark Low-Light Video Enhancement Using Event-Guided Sensor-Level Fusion

Jul 16, 2026

Upload by Rowan

Enhancing videos under extreme low-light conditions is difficult because restoration quality must be balanced with computational efficiency on resource-limited devices. This work presents EeveeDark, a low-light video enhancement framework that fuses the spatial richness of sensor-level RAW data with the temporal precision of event streams. A Binary Neural Network architecture reduces overhead via quantized weights and activations. The model uses modality-specific binary encoders, a lightweight fusion block, and event-guided skip gating for dynamic spatiotemporal refinement, achieving strong results on synthetic and real-world datasets with an improved performance-efficiency trade-off over full-precision baselines.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Introduction

Problem of low-light video enhancement

Event cameras and limitations of existing methods

Method Overview

EeveeDark framework and binary neural architecture

Binary encoders, fusion block, and event-guided skip gating

Experiments and Results

Comparison with state-of-the-art models

EeveeDark: A Binary Neural Framework for Low-Light Video Enhancement via Event-Guided Sensor-Level Fusion Onur Eker1 , Erkut Erdem2 , Senior Member, IEEE, and Aykut Erdem3 , Senior Member, IEEE arXiv :2607 .062 17v 1 [ cs .CV] 7 Jul 2026 Abstract—Enhancing videos under extreme low-light conditions remains challenging due to the difficulty of balancing restoration quality and computational efficiency in resourceconstrained settings. This paper introduces EeveeDark, a lowlight video enhancement framework that combines the spatial richness of sensor-level RAW data with the temporal precision of event streams. Central to our model is a Binary Neural Network (BNN) architecture that reduces computational overhead by quantizing weights and activations while preserving detail. EeveeDark incorporates (i) modality-specific binary encoders for processing RAW frames and event data,(ii) a lightweight fusion block for integrating spatial and temporal cues, and (iii) an event-guided skip gating mechanism for dynamic spatiotemporal refinement. Experiments on synthetic and real-world datasets show that EeveeDark outperforms prior BNN-based methods and offers a favorable performance-efficiency trade-off compared to full-precision models. The project page is available at [https:](https:)//[cyberiada.github.io/EeveeDark/](cyberiada.github.io/EeveeDark/). Index Terms—Sensor fusion, Deep learning for visual perception, Low-light video enhancement, Event camera. I. INTRODUCTION LOW-LIGHT video enhancement is essential for many robotics and vision applications but remains challenging due to severe noise, low contrast, color degradation, and the need for temporal coherence. While recent advances in deep learning have significantly improved image enhancement quality [1], balancing restoration performance with efficiency, particularly on resource-constrained platforms, remains an open problem. Event cameras offer a promising sensing modality for lowlight scenarios due to their high temporal resolution and dynamic range. However, existing event-guided enhancement methods that integrate event streams with RGB images [2]–[5] face two critical limitations: (i) they rely on computationally intensive components (e.g., optical flow or attention-based alignment), hindering deployment on resource-constrained platforms, and (ii) their dependence on processed RGB inputs Manuscript received: November 10, 2025; Revised January 21, 2026; Accepted February 10, 2026 . This paper was recommended for publication by Editor Pascal Vasseur upon evaluation of the Associate Editor and Reviewers’ comments. This work was supported by TUBITAK-1001 Program Award No. 121E454 . 1 Onur Eker is with the Department of Computer Engineering, Hacettepe University, TR-06800 Ankara, Turkey, and also with HAVELSAN Inc., TR- 06510, Ankara, [Turkey.](Turkey. onureker@hacettepe.edu.tr)[ onureker@hacettepe.edu.tr](Turkey. onureker@hacettepe.edu.tr) 2Erkut Erdem is with the Department of Computer Engineering, Hacettepe University, TR-06800 Ankara, Turkey, and also with the Koc University Is Bank AI Center, TR-34450, Istanbul, Turkey. [erkut@cs.hacettepe.edu.tr](erkut@cs.hacettepe.edu.tr) 3Aykut Erdem is with the Department of Computer Engineering, Koc University, TR-34450 Istanbul, Turkey, and also with the Koc University Is Bank AI Center, TR-34450, Istanbul, [Turkey.](Turkey. aerdem@ku.edu.tr)[ aerdem@ku.edu.tr](Turkey. aerdem@ku.edu.tr) Digital Object Identifier (DOI): 10.1109/LRA.2026.3666388 . 33 32 ShiftNet 31 EeveeDark (Ours) (13 .38M) (0.35M) 30 FloRNN EvLight BRVE (10 .49M) (22.73M) (0.30M) 29 BBCU (0.30M) 28 0 10 20 30 40 50 FLOPs (Billions) Fig. 1. Comparison with state-of-the-art models. PSNR versus computational complexity is shown, with circle sizes indicating parameter count (millions) . EeveeDark outperforms prior BNN methods BBCU and BRVE by a significant margin, and offers a favorable performance-efficiency trade-o

FAQs

What is EeveeDark designed to solve in low-light video enhancement?

It targets the difficulty of restoring video quality under extreme darkness while keeping computation efficient enough for resource-constrained deployment.

How does EeveeDark combine different sensor modalities?

It integrates sensor-level RAW data for spatial detail with event streams for temporal precision using a lightweight fusion block and event-guided refinement.

What makes EeveeDark computationally efficient compared with full-precision models?

It uses a Binary Neural Network design that quantizes weights and activations to reduce overhead while preserving detail and temporal consistency.

Jul 16, 2026

Upload by Rowan