Research map / Label-efficient and robust learning / Noisy labels
CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images
Sheng Guo, Weilin Huang, Haozhi Zhang, Chenfan Zhuang, Dengke Dong, Matthew R. Scott, Dinglong Huang
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 the noisy subset as a second curriculum stage (ablation) | CurriculumNet on BN-Inception · WebVision val | top-5 error: 12.98 -> 11.38 | Improved |
| add the highly noisy subset as a third curriculum stage (ablation) | CurriculumNet on BN-Inception · WebVision val | top-5 error: 11.38 -> 10.82 | Improved |
| use the 3-subset curriculum instead of training on all data directly (ablation) | CurriculumNet on BN-Inception · WebVision val | top-1 error: 30.16 -> 27.91 | Improved |
| use the 3-subset curriculum instead of training on all data directly (ablation) | CurriculumNet on BN-Inception · WebVision val | top-5 error: 12.43 -> 10.82 | Improved |
| use the 3-subset curriculum instead of training on all data directly (ablation) | CurriculumNet on BN-Inception · ImageNet val (trained on WebVision) | top-5 error: 16.20 -> 15.11 | Improved |
| train on the clean subset only instead of all data (ablation) | BN-Inception · WebVision val | top-5 error: 12.43 -> 12.98 | Hurt |
| use 50% of the highly noisy subset instead of 0% (ablation) | Model-D · WebVision val | top-5 error: 11.38 -> 10.82 | Improved |
| increase highly noisy data from 50% to 100% (ablation) | Model-D · WebVision val | top-5 error: 10.82 -> 10.94 | No clear effect |
| use density-based clustering instead of K-means for the curriculum (ablation) | CurriculumNet Model-D · WebVision val | top-5 error: 16.6 -> 11.5 | No clear effect |
| use Inception_v3 instead of Inception_v2 (ablation) | CurriculumNet Model-D · WebVision val | top-5 error: 10.82 -> 7.88 | Improved |
| use Inception_resnet_v2 instead of Inception_v3 (ablation) | CurriculumNet Model-D · WebVision val | top-5 error: 7.88 -> 6.38 | Improved |
| add multi-scale 5x5, 7x7, 9x9 kernels in the first conv layer (ablation) | BN-Inception · WebVision val | top-5 error: about 0.5% | Improved |
Later papers that compared with this method
4 later accepted papers report CurriculumNet as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Learning From Web Data With Self-Organizing Memory Module (CVPR 2020 · better on 6 of 6 shared results)
- Suppressing Mislabeled Data via Grouping and Self-Attention (ECCV 2020 · better on 1 of 1 shared result)
- Weakly Supervised Learning with Side Information for Noisy Labeled Images (ECCV 2020 · better on 1 of 1 shared result)
- O2U-Net: A Simple Noisy Label Detection Approach for Deep Neural Networks (ICCV 2019 · better on 8 of 8 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CleanNet — CleanNet: Transfer Learning for Scalable Image Classifier Training With Label Noise (CVPR 2018) · better on 1 of 1 shared result
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
- Attend in Groups: A Weakly-Supervised Deep Learning Framework for Learning From Web Data (CVPR 2017 · similarity 0.89)
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.88)
- Weakly Supervised Learning with Side Information for Noisy Labeled Images (ECCV 2020 · similarity 0.87)
- MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels (ICML 2018 · similarity 0.87)
- Webly-Supervised Fine-Grained Visual Categorization via Deep Domain Adaptation (TPAMI 2016 · similarity 0.87)
- Curriculum Labeling: Revisiting Pseudo-Labeling for Semi-Supervised Learning (AAAI 2021 · 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.
