Research map / Image generation / Image restoration
Q-Norm: Robust Representation Learning via Quality-Adaptive Normalization
Lanning Zhang, Ying Zhou, Fei Gao, Ziyun Li, Maoying Qiao, Jinlan Xu, Nannan Wang
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 |
|---|---|---|---|
| replace BN with channel-wise Q-Norm on CODaN Night | ResNet18 · CODaN Night | Top1 accuracy: 0.308 -> 0.292 | Hurt |
| use spatial-wise Q-Norm instead of BN | ResNet18 · CODaN Night | Top1 accuracy: 0.308 -> 0.356 | Improved |
| add Q-Norm to detector | RT-DETR · HazyDet | Avg mAP50: 45.40 -> 48.00 | Improved |
| add Q-Norm to detector | YOLOv11 · HazyDet | Avg mAP50: 63.00 -> 63.70 | Improved |
| Q-Norm versus data augmentation on blurred VOC | YOLOv11 · VOC blur (avg) | mAP50: 63.30 -> 64.00 | Improved |
| Q-Norm versus data augmentation on blurred PASCAL-Context semseg | MTMamba · PASCAL-Context semseg (avg) | mIoU: 76.40 -> 76.19 (ch), 76.46 (ch-sp) | No clear effect |
| Q-Norm versus data augmentation on normal estimation | MTMamba · PASCAL-Context normals (avg) | mErr (lower better): 14.99 -> 15.16 | Hurt |
| Q-Norm versus data augmentation on image translation | MATEBIT · Metface | FID: 39.56 -> 34.61 | Improved |
| use CONTRIQUE instead of BRISQUE as quality model (ablation) | Q-Norm · CODaN Night | accuracy %: 30.80 -> 35.60 | Improved |
| use any IQA model for Q-Norm on HazyDet (ablation) | Q-Norm · HazyDet | mAP50 %: 45.40 -> 47.60 (BRISQUE), 48.00 (CNN, CONTRIQUE) | Improved |
| apply channel then spatial Q-Norm jointly instead of spatial then channel (ablation) | YOLOv11 · VOC | avg mAP50: 59.85 -> 63.60 | Improved |
| insert Q-Norm at the front of the backbone rather than later layers (ablation) | YOLOv11, MTMamba · VOC / PASCAL-Context | mAP / mIoU | Improved |
Closest papers by meaning
- Convolutional Normalization: Improving Deep Convolutional Network Robustness and Training (NeurIPS 2021 · similarity 0.85)
- QuARF: Quality-Adaptive Receptive Fields for Degraded Image Perception (AAAI 2025 · similarity 0.85)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.84)
- Normalize Filters! Classical Wisdom for Deep Vision (NeurIPS 2025 · similarity 0.84)
- Adaptive Feature Selection for No-Reference Image Quality Assessment by Mitigating Semantic Noise Sensitivity (ICML 2024 · similarity 0.84)
- Q-Insight: Understanding Image Quality via Visual Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
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.
