Research map / Label-efficient and robust learning / Domain adaptation and generalization
Ranking Distance Calibration for Cross-Domain Few-Shot Learning
Pan Li, Shaogang Gong, Chengjie Wang, Yanwei Fu
Check 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 |
|---|---|---|---|
| Calibrate pairwise distances by reciprocal-neighbor re-ranking using task support and query examples | NPC with ResNet10 · 5-way 1-shot, average over 8 target domains | accuracy: 39.72 -> 48.65 | Improved |
| Calibrate pairwise distances by reciprocal-neighbor re-ranking using task support and query examples | NPC with ResNet10 · 5-way 5-shot, average over 8 target domains | accuracy: 57.42 -> 60.06 | Improved |
| Fine-tune the representation by mapping its distance distribution toward calibrated distances | ResNet10 CD-FSL representation · 5-way 1-shot, average over 8 target domains | accuracy: 46.25 -> 51.53 | Improved |
| Fine-tune the representation by mapping its distance distribution toward calibrated distances | ResNet10 CD-FSL representation · 5-way 5-shot, average over 8 target domains | accuracy: 64.68 -> 63.70 | Mixed |
| Add the task-adaptive nonlinear subspace to RDC (ablation) | RDC · 5-way 1-shot, average over 8 target domains | accuracy: 47.71 -> 48.65 | Improved |
| Add the task-adaptive nonlinear subspace to RDC (ablation) | RDC · 5-way 5-shot, average over 8 target domains | accuracy: 58.97 -> 60.06 | Improved |
| Add the task-adaptive nonlinear subspace to RDC-FT (ablation) | RDC-FT · 5-way 1-shot, average over 8 target domains | accuracy: 51.03 -> 51.53 | Improved |
| Add the task-adaptive nonlinear subspace to RDC-FT (ablation) | RDC-FT · 5-way 5-shot, average over 8 target domains | accuracy: 62.91 -> 63.70 | Improved |
| Use a tanh nonlinear subspace rather than the original space or tested KPCA kernels (ablation) | RDC · 5-way 1-shot, average over 8 target domains | accuracy: 47.41 -> 48.36 | Improved |
| Use a tanh nonlinear subspace rather than the original space (ablation) | RDC · 5-way 5-shot, average over 8 target domains | accuracy: 59.09 -> 59.83 | Improved |
| Apply attention weighting to reciprocal-neighbor distances during fine-tuning (ablation) | RDC-FT · 5-way 1-shot, average over 8 target domains | accuracy: 51.09 -> 51.53 | Improved |
| Apply attention weighting to reciprocal-neighbor distances during fine-tuning (ablation) | RDC-FT · 5-way 5-shot, average over 8 target domains | accuracy: 63.16 -> 63.70 | Improved |
Closest papers by meaning
- From Instance to Metric Calibration: A Unified Framework for Open-World Few-Shot Learning (TPAMI 2023 · similarity 0.89)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.88)
- Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.87)
- Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning (ICML 2025 · similarity 0.86)
- The Devil is in Low-Level Features for Cross-Domain Few-Shot Segmentation (CVPR 2025 · similarity 0.86)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · 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.
