Research map / Label-efficient and robust learning / Recognition and long-tail
Weakly Supervised Learning with Side Information for Noisy Labeled Images
Lele Cheng, Xiangzeng Zhou, Liming Zhao, Dangwei Li, Hong Shang, Yun Zheng, Pan Pan, Yinghui Xu
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 |
|---|---|---|---|
| train on only a high-confidence subset instead of all data (ablation) | ResNeXt-101 Model-A · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 51.05% -> 47.81% | Hurt |
| add confidence-score loss reweighting (SINet) | ResNeXt-101 Model-A · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 51.05% -> 55.57% | Improved |
| add confidence-score loss reweighting (SINet) | ResNeXt-101 Model-A · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-5 accuracy: 74.94% -> 78.34% | Improved |
| build class relation graph from category description plus hierarchical WordNet instead of category name only (ablation) | SINet · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 54.18% -> 55.98% | Improved |
| combine description graph with WordNet graph instead of description alone (ablation) | SINet · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 55.63% -> 55.98% | Improved |
| weight prototype candidates by consistency score instead of plain averaging (ablation) | SINet visual prototype · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 55.68% -> 55.98% | Improved |
| shift factor alpha=1.2 instead of 0.8 (ablation) | SINet noise weighting · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 53.25% -> 55.98% | Improved |
| shift factor alpha=1.4 instead of 1.2 (ablation) | SINet noise weighting · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 55.98% -> 55.76% | Hurt |
| contrast factor beta=2.5 instead of 1.5 (ablation) | SINet noise weighting · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 55.98% -> 55.35% | Hurt |
| use SENet154 backbone instead of ResNeXt101 | SINet · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 55.56% -> 56.87% | Improved |
| use SE-ResNeXt101 instead of ResNeXt101 | SINet · WebVision 1.0 val (WebVision 2.0 training, ResNeXt-101) | top-1 accuracy: 55.56% -> 55.57% | No clear effect |
| add SINet to ResNet50 on ImageNet with 40% noise | ResNet50 · ImageNet (40% uniform label noise) | top-1 accuracy: 66.47 -> 69.12 | Improved |
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.
- CurriculumNet — CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images (ECCV 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
- Weakly Supervised Image Classification Through Noise Regularization (CVPR 2019 · similarity 0.88)
- CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images (ECCV 2018 · similarity 0.87)
- Attend in Groups: A Weakly-Supervised Deep Learning Framework for Learning From Web Data (CVPR 2017 · similarity 0.86)
- Learning From Large-Scale Noisy Web Data With Ubiquitous Reweighting for Image Classification (TPAMI 2019 · similarity 0.85)
- Exploiting Negative Evidence for Deep Latent Structured Models (TPAMI 2018 · similarity 0.85)
- Fusing Conditional Submodular GAN and Programmatic Weak Supervision (AAAI 2024 · 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.
