Research map / Label-efficient and robust learning / Noisy labels
MetaCleaner: Learning to Hallucinate Clean Representations for Noisy-Labeled Visual Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 MetaCleaner to the standard CNN training procedure | ResNet50 · Food-101N | accuracy: 81.44 -> 82.52 | Helped |
| Add MetaCleaner to the standard CNN training procedure | ResNet50 · Clothing1M | accuracy: 68.94 -> 72.50 | Helped |
| Use MetaCleaner rather than the strongest listed baseline | Food-101N · Food-101N | accuracy: 83.95 -> 85.05 | Helped |
| Use MetaCleaner rather than JointOptim | Clothing1M, Noise1M · Clothing1M | accuracy: 72.23 -> 72.50 | Helped |
| Use MetaCleaner instead of Reweight with a clean training set | Cifar-10 with 40% uniform-flip noise · Cifar-10 | accuracy: 86.92 -> 90.09 | Helped |
| Add MetaCleaner during training | ResNet50 · ImageNet with 40% uniform-flip noise | top-1 accuracy: 66.47 -> 69.12 | Helped |
| Estimate image importance with learned weights rather than constant weights (ablation) | MetaCleaner · Food-101N | accuracy: 81.24 -> 82.52 | Helped |
| Estimate image importance with learned weights rather than constant weights (ablation) | MetaCleaner · Clothing1M | accuracy: 70.41 -> 72.50 | Helped |
| Use subset-level FC weighting instead of individual-image attention (ablation) | MetaCleaner · Food-101N | accuracy: 81.44 -> 82.52 | Helped |
| Use subset-level FC weighting instead of individual-image attention (ablation) | MetaCleaner · Clothing1M | accuracy: 71.08 -> 72.50 | Helped |
| Add ReLU in the first FC layer of the weighting network (ablation) | MetaCleaner · Food-101N | accuracy: 82.09 -> 82.52 | Helped |
| Add ReLU in the first FC layer of the weighting network (ablation) | MetaCleaner · Clothing1M | accuracy: 71.15 -> 72.50 | Helped |
Closest papers by meaning
- Deep Neural Networks Learn Meta-Structures from Noisy Labels in Semantic Segmentation (AAAI 2022 · similarity 0.86)
- CleanNet: Transfer Learning for Scalable Image Classifier Training With Label Noise (CVPR 2018 · similarity 0.86)
- Low-Shot Learning From Imaginary Data (CVPR 2018 · similarity 0.85)
- Learning to Learn From Noisy Labeled Data (CVPR 2019 · similarity 0.85)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.84)
- Analyzing and Mitigating Object Hallucination: A Training Bias Perspective (AAAI 2026 · similarity 0.83)
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.