Research map / Label-efficient and robust learning / Semi-supervised learning
Towards Good Practices for Efficiently Annotating Large-Scale Image Classification Datasets
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 |
|---|---|---|---|
| Use the proposed framework instead of Lean | Online image annotation on ImageNet100 · ImageNet100 | top-1 label accuracy: 79.5\% -> 87.4\% | Helped |
| Use the proposed framework instead of Online DS | Online image annotation on ImageNet100 · ImageNet100 | top-1 label accuracy: 46.9\% -> 87.4\% | Helped |
| Add self-supervised features to online labeling | Online labeling across ImageNet subsets · ImageNet subsets | annotation efficiency: up to 5x improvement at similar accuracy | Helped |
| Use BYOL features rather than no learning model in the loop | Online labeling on Dog · Dog | annotations required at similar accuracy: 5x fewer annotations | Helped |
| Add pseudo-labeling | Online labeling across ImageNet subsets · ImageNet subsets | annotation efficiency | Helped |
| Use pseudo-labeling rather than modified MixMatch (ablation) | Online labeling with fixed features · ImageNet subsets | annotation efficiency | Helped |
| Add pseudo-labeling (ablation) | Online labeling on Dog · Dog | annotations at convergence: 30% fewer annotations | Helped |
| Use soft EM convergence rather than the hard constraint (ablation) | Online labeling with semi-supervised learning · ImageNet subsets | stability of label quality | Helped |
| Keep the best model across time steps using prototype validation (ablation) | Online labeling with semi-supervised learning · ImageNet subsets | stability of label quality | Helped |
| Use a clean prototype validation set instead of noisy cross-validation splits (ablation) | Model likelihood calibration in online labeling · ImageNet subsets | annotation efficiency | Helped |
| Use fewer workers (ablation) | Online labeling on Dog · Dog | accuracy: 17% | Helped |
| Use fewer workers (ablation) | Online labeling across datasets · ImageNet subsets | annotation efficiency | Mixed |
Closest papers by meaning
- Is one annotation enough? - A data-centric image classification benchmark for noisy and ambiguous label estimation (NeurIPS 2022 · similarity 0.91)
- Exploiting weakly supervised visual patterns to learn from partial annotations (NeurIPS 2020 · similarity 0.89)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.89)
- Lean Multiclass Crowdsourcing (CVPR 2018 · similarity 0.89)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · similarity 0.88)
- Modeling Collaborator: Enabling Subjective Vision Classification With Minimal Human Effort via LLM Tool-Use (CVPR 2024 · 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.