Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Revisiting Pose-Normalization for Fine-Grained Few-Shot Recognition
Check 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 |
|---|---|---|---|
| Add pose normalization to transfer learning | Transfer learning with 4-layer ConvNet · CUB | all-shot accuracy: 33.42 -> 49.96 | Helped |
| Add pose normalization to prototypical networks | Proto with ResNet18 · CUB | all-shot accuracy: 42.73 -> 63.44 | Helped |
| Add pose normalization to dynamic few-shot learning | Dynamic with 4-layer ConvNet · CUB | all-shot accuracy: 35.77 -> 54.17 | Helped |
| Use ground-truth pose instead of estimated pose (ablation) | Proto+PN with 4-layer ConvNet · CUB | all-shot accuracy: 49.56 -> 59.55 | Helped |
| Use ground-truth pose instead of estimated pose (ablation) | Proto+PN with ResNet18 · CUB | all-shot accuracy: 63.44 -> 62.63 | Hurt |
| Use ground-truth pose instead of estimated pose (ablation) | Dynamic+PN with 4-layer ConvNet · CUB | all-shot accuracy: 54.17 -> 62.67 | Helped |
| Use ground-truth pose instead of estimated pose (ablation) | Dynamic+PN with ResNet18 · CUB | all-shot accuracy: 60.19 -> 60.09 | Hurt |
| Train pose estimator using annotations on fewer images (ablation) | Proto+PN with 4-layer ConvNet · CUB | all-shot accuracy: 40.66 (5%); 50.51 (100%) | Mixed |
| Train pose estimator using annotations on fewer images (ablation) | Proto+PN with ResNet18 · CUB | all-shot accuracy: 52.78 (5%); 66.33 (100%) | Mixed |
| Add pose normalization | Proto with 4-layer ConvNet · NA | mean accuracy: 8.73 -> 21.02 | Helped |
| Add pose normalization | Proto with ResNet18 · NA | per-class accuracy: 12.55 -> 30.59 | Helped |
| Add pose normalization | Proto with 4-layer ConvNet · FGVC-Aircraft | all-shot accuracy: 52.06 -> 60.83 | Helped |
Closest papers by meaning
- Fine-Grained Object Classification via Self-Supervised Pose Alignment (CVPR 2022 · similarity 0.87)
- Rethinking Generalization in Few-Shot Classification (NeurIPS 2022 · similarity 0.87)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · similarity 0.86)
- Few-Shot Learning With Localization in Realistic Settings (CVPR 2019 · similarity 0.85)
- Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement (ICLR 2025 · similarity 0.84)
- Simple Semantic-Aided Few-Shot Learning (CVPR 2024 · similarity 0.84)
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.