Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Few-Shot Image Recognition by Predicting Parameters From Activations
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 |
|---|---|---|---|
| Predict novel-class parameters from activation statistics with Ours-phi2 instead of fine-tuning | ResNet-50 on ImageNet · ImageNet (1000-way), 1-shot, C_few | Top-1 accuracy: 2.90% -> 21.58% | Helped |
| Predict novel-class parameters from activation statistics with Ours-phi2 instead of fine-tuning | ResNet-50 on ImageNet · ImageNet (1000-way), 2-shot, C_few | Top-1 accuracy: 11.54% -> 27.44% | Helped |
| Predict novel-class parameters from activation statistics with Ours-phi2 instead of fine-tuning | ResNet-50 on ImageNet · ImageNet (1000-way), 3-shot, C_few | Top-1 accuracy: 18.56% -> 31.20% | Helped |
| Use a two-layer predictor with a ReLU nonlinearity rather than a one-layer predictor (ablation) | Ours on ImageNet · ImageNet (1000-way), C_few, 1/2/3-shot | Top-1 accuracy: 19.88% -> 21.58%; 25.54% -> 27.44%; 28.72% -> 31.20% | Helped |
| Use a two-layer predictor with a ReLU nonlinearity rather than a one-layer predictor (ablation) | Ours on ImageNet · ImageNet (1000-way), C_large, 1/2/3-shot | Top-1 accuracy: 72.56% -> 74.17%; 71.94% -> 73.43%; 71.56% -> 72.98% | Mixed |
| Train the predictor with individual activations as well as mean activations, instead of mean activations only (ablation) | Ours on ImageNet · ImageNet (1000-way), C_few, 1/2/3-shot | Top-1 accuracy: 14.32% -> 21.58%; 18.70% -> 27.44%; 22.32% -> 31.20% | Helped |
| Train the predictor with individual activations as well as mean activations, instead of mean activations only (ablation) | Ours on ImageNet · ImageNet (1000-way), C_large, 1/2/3-shot | Top-1 accuracy: 75.63% -> 74.17%; 75.44% -> 73.43%; 75.34% -> 72.98% | Hurt |
| Use Ours-WRN rather than MAML | MiniImageNet · 5-way 1-shot | Accuracy: 48.70 ± 1.84% -> 59.60 ± 0.41% | Helped |
| Use Ours-WRN rather than MAML | MiniImageNet · 5-way 5-shot | Accuracy: 63.11 ± 0.92% -> 73.74 ± 0.19% | Helped |
Closest papers by meaning
- Dynamic Few-Shot Visual Learning Without Forgetting (CVPR 2018 · similarity 0.92)
- Dense Classification and Implanting for Few-Shot Learning (CVPR 2019 · similarity 0.91)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · similarity 0.90)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.90)
- Self-Prompt Mechanism for Few-Shot Image Recognition (AAAI 2024 · similarity 0.88)
- 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.