Research map / Label-efficient and robust learning / Recognition and long-tail
Labeling the Features Not the Samples: Efficient Video Classification with Minimal Supervision
Marius Leordeanu, Alexandra Radu, Shumeet Baluja, Rahul Sukthankar
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 unlabeled test-set frames to the training data (ablation) | Almost unsupervised feature selection · YouTube-Objects, Features I+II+III, 8 labeled shots/class | testing accuracy: 53.71% -> 78.05% | Improved |
| Add unlabeled test-set frames (ablation) | Almost unsupervised feature selection · YouTube-Objects, Features III, 8 labeled shots/class | testing accuracy: 49.03% -> 74.30% | Improved |
| Use all three feature types instead of Feature I alone (ablation) | Almost unsupervised feature selection · YouTube-Objects, 8 labeled shots/class, Train + 100% test S_A | testing accuracy: 43.00% -> 78.05% | Improved |
| Use all three feature types instead of Feature I alone (ablation) | Almost unsupervised feature selection · YouTube-Objects, 1 labeled shot/class | improvement over supervised method: +15.1 -> +24.4 | Improved |
| Use almost unsupervised learning with unlabeled data rather than supervised learning | Feature selection and classifier learning · YouTube-Objects, Feature I+II+III, 1 labeled shot | improvement over supervised method: +24.4 | Improved |
| Use almost unsupervised learning with unlabeled data rather than supervised learning | Feature selection and classifier learning · YouTube-Objects, Feature I+II+III, 16 labeled shots | improvement over supervised method: +5.4 | Improved |
| Estimate feature signs from limited labeled examples | Almost unsupervised feature selection · YouTube-Objects | accuracy | Improved |
| Increase the number of selected features k (ablation) | Proposed feature selection method · YouTube-Objects | accuracy | Mixed |
| Add more unlabeled frames (ablation) | Almost unsupervised feature selection · YouTube-Objects | testing accuracy | Improved |
| Use Elastic Net instead of Lasso | Feature selection baseline · YouTube-Objects | performance relative to Lasso: at best, 2% superior | No clear effect |
| Use the proposed feature-average classifier rather than SVM with selected features | Proposed feature selection method · YouTube-Objects, limited labeled training data | generalization | Improved |
Closest papers by meaning
- Joint Representative Selection and Feature Learning: A Semi-Supervised Approach (CVPR 2019 · similarity 0.87)
- Submodular Attribute Selection for Visual Recognition (TPAMI 2016 · similarity 0.87)
- Unsupervised Object Segmentation in Video by Efficient Selection of Highly Probable Positive Features (ICCV 2017 · similarity 0.87)
- Unsupervised Feature Learning via Non-Parametric Instance Discrimination (CVPR 2018 · similarity 0.87)
- SAND: One-Shot Feature Selection with Additive Noise Distortion (ICML 2025 · similarity 0.86)
- Enhanced Sample Selection with Confidence Tracking: Identifying Correctly Labeled Yet Hard-to-Learn Samples in Noisy Data (AAAI 2025 · similarity 0.86)
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.
