Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn feature-wise transformations using multiple seen domains instead of fixed FT parameters | MatchingNet · CUB, Cars, Places, Plantae, 5-way 5-shot leave-one-out | accuracy: 56.29 -> 61.41; 39.58 -> 43.08; 62.32 -> 64.99; 46.48 -> 48.32 | Helped |
| Learn feature-wise transformations using multiple seen domains instead of fixed FT parameters | RelationNet · CUB, Cars, Places, Plantae, 5-way 5-shot leave-one-out | accuracy: 63.64 -> 64.99; 42.24 -> 43.44; 65.42 -> 67.35; 47.81 -> 50.39 | Helped |
| Learn feature-wise transformations using multiple seen domains instead of fixed FT parameters | GNN · CUB, Cars, Places, Plantae, 5-way 5-shot leave-one-out | accuracy: 70.37 -> 73.11; 47.68 -> 49.88; 74.48 -> 77.05; 57.85 -> 58.84 | Helped |
| Add feature-wise transformation layers with fixed parameters | GNN · CUB, 5-way 5-shot, mini-ImageNet training | accuracy: 62.25 -> 66.98 | Helped |
| Add feature-wise transformation layers with fixed parameters | GNN · Places, 5-way 5-shot, mini-ImageNet training | accuracy: 70.84 -> 73.94 | Helped |
| Set transformation hyperparameters to {1,1} instead of {0.3,0.5} (ablation) | MatchingNet · CUB, 5-way 1-shot | accuracy: 36.61 -> 31.74 | Hurt |
| Set transformation hyperparameters to {1,1} instead of {0.3,0.5} (ablation) | GNN · Plantae, 5-way 1-shot | accuracy: 35.95 -> 36.73 | Mixed |
| Pre-train the feature encoder on mini-ImageNet categories (ablation) | GNN · CUB, 5-way 1-shot leave-one-out | accuracy: 37.21 -> 49.46 | Helped |
| Use the learning-to-learn procedure with only one seen domain (ablation) | RelationNet · CUB, Cars, Places, Plantae, 5-way 5-shot | accuracy: 59.46 -> 58.19; 39.91 -> 39.35; 66.28 -> 66.17; 45.08 -> 46.75 | Hurt |
| Initialize learned transformation parameters to {0.1,0.3} instead of {0.3,0.5} (ablation) | RelationNet · CUB, 5-way 5-shot | accuracy: 64.99 ± 0.54% vs 64.79±0.55% | No clear effect |
Closest papers by meaning
- CAD: Co-Adapting Discriminative Features for Improved Few-Shot Classification (CVPR 2022 · similarity 0.90)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.90)
- A Universal Representation Transformer Layer for Few-Shot Image Classification (ICLR 2021 · similarity 0.89)
- Adaptive Cross-Modal Few-shot Learning (NeurIPS 2019 · similarity 0.89)
- ProD: Prompting-To-Disentangle Domain Knowledge for Cross-Domain Few-Shot Image Classification (CVPR 2023 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
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.