---
license: mit
---
<h2 align="center">
  <b>From Enhancement to Understanding: Build a Generalized Bridge for Low-light Vision via Semantically Consistent Unsupervised Fine-tuning</b>

  <b><i> ICCV 2025</i></b>
    [<a href="https://arxiv.org/abs/2507.08380">arXiv</a>]
</h2>

<div align=center>
<img src="Assets/intro.jpg" width="400px">
</div>

## Abstract
Low-level enhancement and high-level visual understanding in low-light vision have traditionally been treated separately. Low-light enhancement improves image quality for downstream tasks but has limited generalization. Low-light visual understanding, constrained by scarce labeled data, primarily relies on task-specific domain adaptation, which lacks scalability. To address these challenges, we build a generalized bridge between low-light enhancement and low-light understanding, which we term **Generalized Enhancement For Understanding (GEFU)**. This paradigm improves both **generalization** and **scalability**. To tackle the diverse causes of low-light degradation, we propose **Semantically Consistent Unsupervised Fine-tuning (SCUF)**. Extensive experiments demonstrate that our proposed method outperforms current state-of-the-art approaches in terms of traditional image quality as well as GEFU tasks, including classification, detection, and semantic segmentation.

Please see our paper and [github](https://github.com/wangsen99/GEFU) for details.