Abstract

Enhancing videos under extreme low-light conditions remains challenging due to the difficulty of balancing restoration quality and computational efficiency in resource-constrained settings. This paper introduces EeveeDark, a low-light 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.

Method Overview

Overview of EeveeDark. It comprises five key components: (1) Preprocessing and modality-specific encoding, where Bayer RAW frames and event data are separately encoded using binary encoders; (2) an Efficient Multi-Modal Fusion Block that integrates the spatial detail of RAW inputs with the temporal precision of event streams; (3) a Shift Encoder that propagates temporal information across frames and modulates features through Event-Guided Skip Gates (EGSG); (4) a Shift Decoder that refines and aligns features across temporal windows; and (5) a Restoration Module that reconstructs clean, temporally coherent enhanced RAW frames, which are subsequently converted to RGB format via a standard Image Signal Processor (ISP).

Performance-Complexity Trade-off

EeveeDark achieves a compelling balance between restoration quality (PSNR) and computational efficiency (FLOPs).

Compared to BNN-based methods (BBCU, BRVE), our approach significantly improves visual quality while maintaining binary efficiency.

Against full-precision models like ShiftNet and EvLight, EeveeDark offers competitive performance with dramatically reduced computational demands—making it suitable for resource-constrained robotic platforms.

Qualitative Results

Low Light Input Events BRVE EeveeDark (Ours) Reference

Qualitative comparisons on SDE and SDSD RGB benchmarks. EeveeDark consistently outperforms the binary baseline BRVE: it restores finer structural details and improved local contrast, whereas BRVE tends to oversmooth edges and can amplify noise in dark regions. The improvement stems from effectively leveraging event data, which provides high-frequency temporal cues absent in RGB inputs. EeveeDark also exhibits superior frame-to-frame consistency and avoids the flickering artifacts observed in BRVE, while offering a compelling performance–efficiency trade-off at a fraction of the cost of full-precision models.

BibTeX

@article{eker2026eeveedark,
  title={EeveeDark: A Binary Neural Framework for Low-Light Video Enhancement via Event-Guided Sensor-Level Fusion},
  author={Eker, Onur and Erdem, Erkut and Erdem, Aykut},
  journal={IEEE Robotics and Automation Letters},
  year={2026},
  publisher={IEEE}
}

Acknowledgements

This work was supported by TUBITAK-1001 Program Award No. 121E454.

For questions, please contact onureker@hacettepe.edu.tr