Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Dual Adaptive Representation Alignment for Cross-Domain Few-Shot Learning
Yifan Zhao, Tong Zhang, Jia Li, Yonghong Tian
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 prototypical feature alignment (PFA) to Ridge baseline (ablation) | Dara · CropDiseases 5-way 5-shot | accuracy: 73.83 -> 93.55 | Improved |
| add normalized distribution alignment (mean statistics) to Ridge baseline (ablation) | Dara · EuroSAT 5-way 5-shot | accuracy: 63.18 -> 81.25 | Improved |
| add NDA on top of Ridge+PFA (ablation) | Dara · ISIC 5-way 5-shot | accuracy: 51.70 -> 56.28 | Improved |
| use query statistics instead of support statistics for normalization (BN) (ablation) | Dara NDA · CropDiseases 5-way 5-shot | accuracy: 93.55 -> 94.49 | Improved |
| use query statistics instead of support statistics for normalization (BN) (ablation) | Dara NDA · ISIC 5-way 5-shot | accuracy: 51.69 -> 51.29 | Hurt |
| replace mean fusion (0.5,0.5) with learnable gating (ablation) | Dara NDA · ISIC 5-way 5-shot | accuracy: 53.84 -> 56.28 | Improved |
| add instance recalibration to prototypical reprojection (ablation) | Dara PFA · ISIC 5-way 5-shot | accuracy: 55.40 -> 56.28 | Improved |
| use only one query sample for statistics (S+Q-1) instead of all queries (ablation) | Dara NDA · ISIC 5-way 5-shot | accuracy: 55.12 vs 56.28 | No clear effect |
| optimize only reprojection weights Z from pretrained init (ablation) | Dara meta-finetuning · CropDiseases 5-way 5-shot | accuracy: 73.83 -> 94.42 | Improved |
| add backbone optimization to Z optimization (two-stage) (ablation) | Dara meta-finetuning · EuroSAT 5-way 5-shot | accuracy: 83.94 -> 85.84 | Improved |
| use smaller pseudo support (5% support / 95% query) split instead of 20/80 (query/support) (ablation) | Dara meta-finetuning · ISIC 5-way 20-shot | accuracy: 69.12 -> 67.36 | Hurt |
| increase feature resolution from 3x3 to 7x7 (ablation) | Dara PFA · ChestX 5-way 5-shot | accuracy: 27.57 -> 26.00 | Hurt |
Closest papers by meaning
- Cross-Domain Cross-Set Few-Shot Learning via Learning Compact and Aligned Representations (ECCV 2022 · similarity 0.92)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.90)
- Meta-Exploiting Frequency Prior for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.88)
- Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining (CVPR 2024 · similarity 0.87)
- Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.87)
- Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (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.
