Research map / Label-efficient and robust learning / Noisy labels
O2U-Net: A Simple Noisy Label Detection Approach for Deep Neural Networks
Jinchi Huang, Lie Qu, Rongfei Jia, Binqiang Zhao
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 |
|---|---|---|---|
| use cyclical learning rate instead of constant learning rate for loss tracking (ablation) | O2U-net detection · CIFAR-10 40% noise, ResNet-101 | detection precision (%): 39.96% -> 95.67% | Improved |
| use cyclical learning rate instead of constant for loss ranking (ablation) | O2U-net detection · CIFAR-100 40% noise, ResNet-101 | detection precision (%): 40.41% -> 92.64% | Improved |
| remove detected top-k noisy labels before retraining | O2U-net · CIFAR-10 80% noise, ResNet-101 | accuracy (%): 15.91% -> 37.76% | Improved |
| use O2U-net instead of Co-Teaching for final classifier | ResNet-101 · Clothing1M | accuracy (%): 78.52% -> 82.38% | Improved |
| use O2U-net instead of CurriculumNet for final classifier | 9-Layer CNN · Clothing1M | accuracy (%): 73.33% -> 75.61% | Improved |
| use O2U-net instead of MentorNet DD | ResNet-101 · CIFAR-10 40% noise | accuracy (%): 88.64% -> 90.33% | Improved |
| use cycle length 50 instead of 10 (ablation) | O2U-net · CIFAR-100 80% noise, ResNet-101 | accuracy (%): 39.39% -> 26.96% | Hurt |
| use cycle length 50 instead of 10 (ablation) | O2U-net · CIFAR-10 80% noise, ResNet-101 | accuracy (%): 37.76% -> 43.41% | Improved |
| vary cycle length (ablation) | O2U-net · CIFAR-100 detection | precision | No clear effect |
| reduce cyclical-training batch size from 128 to 16 (ablation) | O2U-net · 20% random noise, ResNet-101 | detection PR curve | Improved |
| reduce cyclical-training batch size to 4 (ablation) | O2U-net · 20% random noise, ResNet-101 | detection PR curve | Hurt |
| use O2U-net instead of Co-Teaching for detection | ResNet-101 · CIFAR-100 40% noise | detection precision (%): 78.60% -> 92.64% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Co-Teaching — Co-teaching: Robust training of deep neural networks with extremely noisy labels (NeurIPS 2018) · better on 8 of 8 shared results
- CurriculumNet — CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images (ECCV 2018) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Distilling Effective Supervision From Severe Label Noise (CVPR 2020 · similarity 0.90)
- Coresets for Robust Training of Deep Neural Networks against Noisy Labels (NeurIPS 2020 · similarity 0.89)
- Joint Optimization Framework for Learning With Noisy Labels (CVPR 2018 · similarity 0.89)
- Making Deep Neural Networks Robust to Label Noise: A Loss Correction Approach (CVPR 2017 · similarity 0.89)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.89)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
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.
