From Enhancement to Understanding: Build a Generalized Bridge for Low-light Vision via Semantically Consistent Unsupervised Fine-tuning
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 for details.
