Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Task-Level Self-Supervision for Cross-Domain Few-Shot Learning
Wang Yuan, Zhizhong Zhang, Cong Wang, Haichuan Song, Yuan Xie, Lizhuang Ma
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 |
|---|---|---|---|
| Use TL-SS with task variants, consistency and metric-alignment losses alongside HAE (ablation) | T3S · miniImageNet→CUB, 5-way 5-shot | accuracy: 64.87 -> 69.16 | Improved |
| Add instance-level augmentation and metric-alignment loss (ablation) | HAE · miniImageNet→CUB, 5-way 5-shot | accuracy: 64.87 -> 66.55 | Improved |
| Add class-level permutation to instance augmentation (ablation) | HAE with metric-alignment loss · miniImageNet→CUB, 5-way 5-shot | accuracy: 66.55 -> 68.30 | Improved |
| Add task consistency loss Lt (ablation) | HAE with instance augmentation, class permutation and Lm · miniImageNet→CUB, 5-way 5-shot | accuracy: 68.30 -> 69.16 | Improved |
| Replace average task-context pooling with 3D CNN context modeling (ablation) | HAE · miniImageNet→CUB, 5-way 5-shot | accuracy: 63.41 -> 68.30 | Improved |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→CUB, 5-way 1-shot | accuracy: 47.47 -> 45.92 | Hurt |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→Cars, 5-way 1-shot | accuracy: 31.61 -> 33.22 | Improved |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→Places, 5-way 1-shot | accuracy: 55.77 -> 55.83 | Improved |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→EuroSAT, 5-way 1-shot | accuracy: 64.00 -> 65.73 | Improved |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→CUB, 5-way 5-shot | accuracy: 64.97 -> 69.16 | Improved |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→Cars, 5-way 5-shot | accuracy: 47.77 -> 49.82 | Improved |
| Use T3S rather than the strongest listed baseline | Cross-domain few-shot classification · miniImageNet→Places, 5-way 5-shot | accuracy: 71.82 -> 76.33 | Improved |
Closest papers by meaning
- Self-training For Few-shot Transfer Across Extreme Task Differences (ICLR 2021 · similarity 0.90)
- Boosting Few-Shot Visual Learning With Self-Supervision (ICCV 2019 · similarity 0.89)
- Cross-Domain Few-Shot Learning With Task-Specific Adapters (CVPR 2022 · similarity 0.89)
- Task-Adaptive Prompted Transformer for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.88)
- One Meta-tuned Transformer is What You Need for Few-shot Learning (ICML 2024 · similarity 0.87)
- Probing the Mid-level Vision Capabilities of Self-Supervised Learning (CVPR 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.
