Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
A Broader Study of Cross-Domain Few-Shot Learning
Yunhui Guo, Noel C. Codella, Leonid Karlinsky, James V. Codella, John R. Smith, Kate Saenko, Tajana Rosing, Rogerio Feris
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 ProtoNet instead of MetaOpt (older vs newer meta-learner) | meta-learning on BSCD-FSL · BSCD-FSL average | accuracy: 59.78 vs 57.35 | Improved |
| add feature-wise transform (FWT) | ProtoNet · BSCD-FSL average | accuracy: 59.78 -> 56.72 | Hurt |
| add feature-wise transform (FWT) | MatchingNet · BSCD-FSL average | accuracy: 50.21 -> 46.55 | Hurt |
| add feature-wise transform (FWT) | RelationNet · BSCD-FSL average | accuracy: 54.48 -> 52.6 | Hurt |
| replace meta-learning with simple fine-tuning | ProtoNet · BSCD-FSL average | accuracy: 59.78 -> 67.23 | Improved |
| use MatchingNet instead of random embedding | fixed feature extractor · BSCD-FSL average | accuracy: 53.99 vs 50.21 | Hurt |
| add transductive batch statistics to fine-tuning | fine-tune all layers · BSCD-FSL average | accuracy: 67.23 -> 68.14 | Improved |
| use pretrained ImageNet-like weights instead of random weights | fixed feature extractor · BSCD-FSL average | accuracy: 53.99 -> 64.24 | Improved |
| use mean-centroid classifier instead of linear classifier (ablation) | fine-tuning · ISIC 5-way 50-shot | accuracy: 66.48 -> 61.57 | Hurt |
| use cosine-similarity classifier instead of linear classifier (ablation) | fine-tuning · EuroSAT 5-way 50-shot | accuracy: 91.34 -> 88.83 | Hurt |
| use cosine-similarity classifier instead of linear classifier (ablation) | fine-tuning · ChestX 5-way 5-shot | accuracy: 25.97 -> 26.95 | Improved |
| use incremental multi-model selection instead of concatenating all embeddings (ablation) | multiple pretrained models · BSCD-FSL average | accuracy: 68.22 -> 68.69 | Improved |
Closest papers by meaning
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.90)
- A Unified Few-Shot Classification Benchmark to Compare Transfer and Meta Learning Approaches (NeurIPS 2021 · similarity 0.90)
- Meta-Exploiting Frequency Prior for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.89)
- The Devil is in Low-Level Features for Cross-Domain Few-Shot Segmentation (CVPR 2025 · similarity 0.88)
- Adapt Before Comparison: A New Perspective on Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.87)
- A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning (NeurIPS 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.
