Research map / Label-efficient and robust learning / Recognition and long-tail
Weakly Supervised Representation Learning With Coarse Labels
Yuanhong Xu, Qi Qian, Hao Li, Rong Jin, Juhua Hu
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 instance classification to coarse-class training (CoIns vs Cos) | ResNet-18 · CIFAR-100 (100 target classes) | Recall@1: 37.1 -> 57.0 | Improved |
| add coarse-class classification to instance classification (CoIns vs Ins) | ResNet-18 · CIFAR-100 (100 target classes) | Recall@1: 13.6 -> 57.0 | Improved |
| add instance classification to coarse-class training (CoIns vs Cos) | ResNet-18 · SOP (13,160 target classes) | Recall@1: 21.8 -> 35.8 | Improved |
| add instance classification to coarse labels (CoIns vs Cos) | ResNet-50 MoCo-v2 · ImageNet (1000 target classes) | Recall@1: 21.1 -> 51.4 | Improved |
| add coarse labels to MoCo-v2 instance classification | ResNet-50 · ImageNet (1000 target classes, linear probe) | Top-1 accuracy: 67.5 -> 70.4 | Improved |
| add instance classification for coarse-class accuracy | ResNet-18 · CIFAR-100 (20 coarse classes) | Top-1 accuracy: 85.6 -> 86.3 | Improved |
| restrict instance classification within each coarse class (CoIns_imp vs CoIns) | ResNet-18 · CIFAR-100 (100 target classes) | Recall@1: 57.0 -> 56.6 | No clear effect |
| add cosine softmax | CoIns · CIFAR-100 (100 target classes) | Recall@1: 57.0 -> 60.8 | Improved |
| add MLP head on top of cosine softmax | CoIns* · CIFAR-100 (100 target classes) | Recall@1: 60.8 -> 60.5 | No clear effect |
| add MLP head on top of cosine softmax | CoIns* · SOP (13,160 target classes) | Recall@1: 38.1 -> 42.7 | Improved |
| add strong augmentation | CoIns · CIFAR-100 | Recall | Hurt |
| add instance proxy loss at epoch 100 (ablation) | CoIns** · CIFAR-100 (100 target classes) | Recall@1: 60.5 -> 62.0 | Improved |
Closest papers by meaning
- A Weakly Supervised Fine Label Classifier Enhanced by Coarse Supervision (ICCV 2019 · similarity 0.87)
- Grafit: Learning Fine-Grained Image Representations With Coarse Labels (ICCV 2021 · similarity 0.86)
- Coarsely-Labeled Data for Better Few-Shot Transfer (ICCV 2021 · similarity 0.86)
- Embedding Label Structures for Fine-Grained Feature Representation (CVPR 2016 · similarity 0.85)
- Fine-grained Classes and How to Find Them (ICML 2024 · similarity 0.84)
- Twofold Debiasing Enhances Fine-Grained Learning with Coarse Labels (AAAI 2025 · 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.
