Research map / Detection and segmentation / Open-set and few-shot detection
From Enhancement to Understanding: Build a Generalized Bridge for Low-light Vision via Semantically Consistent Unsupervised Fine-tuning
Sen Wang, Shao Zeng, Tianjun Gu, Zhizhong Zhang, Ruixin Zhang, Shouhong Ding, Jingyun Zhang, Jun Wang, Xin Tan, Yuan Xie, Lizhuang Ma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add illumination-aware image prompt to the baseline (ablation) | CycleGAN-Turbo baseline · CODaN top-1 accuracy | Top-1 accuracy (%): 57.24 -> 59.40 | Improved |
| add caption consistency on top of illumination-aware prompt (ablation) | SCUF · LSRW image quality | PSNR (dB): 17.99 -> 16.98 | Hurt |
| add reflectance consistency with decoder-side placement (ablation) | SCUF · LSRW image quality | PSNR (dB): 16.98 -> 18.48 | Improved |
| use cycle-attention adapter instead of IP-Adapter (ablation) | SCUF · CODaN top-1 accuracy | Top-1 accuracy (%): 59.84 -> 60.92 | Improved |
| add caption consistency to full model (ablation) | SCUF · DARK FACE detection | mAP (%): 15.7 -> 16.9 | Improved |
| feed caption prompt directly to the text encoder instead of caption consistency (ablation) | SCUF · CODaN top-1 accuracy | Top-1 accuracy (%): 60.92 -> 60.08 | Hurt |
| add caption prompt to the plain baseline (ablation) | CycleGAN-Turbo baseline · LSRW image quality | PSNR (dB): 14.42 -> 17.27 | Improved |
| enhance low-light images with SCUF before downstream models | RefineNet on BDD100k-night · BDD100k-night | mIoU (%): 11.4 -> 20.1 | Improved |
| use SCUF instead of LightenDiffusion for image quality on LOL | LOL (v2-real) · LOL (v2-real) | PSNR (dB): 22.79 -> 21.32 | Hurt |
| improve image quality metrics of enhancers | low-light enhancement methods · downstream tasks | downstream accuracy | No clear effect |
| use SCUF instead of QuadPrior/LightDiff diffusion enhancers for efficiency | pretrained diffusion enhancers · model complexity | inference time (s): 2.443 -> 0.348 | Improved |
Closest papers by meaning
- Semantically Contrastive Learning for Low-Light Image Enhancement (AAAI 2022 · similarity 0.92)
- Learning Semantic-Aware Knowledge Guidance for Low-Light Image Enhancement (CVPR 2023 · similarity 0.91)
- Improving Visual and Downstream Performance of Low-Light Enhancer with Vision Foundation Models Collaboration (CVPR 2025 · similarity 0.88)
- Luminance-Aware Statistical Quantization: Unsupervised Hierarchical Learning for Illumination Enhancement (NeurIPS 2025 · similarity 0.88)
- Zero-Reference Low-Light Enhancement via Physical Quadruple Priors (CVPR 2024 · similarity 0.87)
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
