Research map / Architectures and efficiency / Attention and transformers
Self-Promoted Supervision for Few-Shot Transformer
Bowen Dong, Pan Zhou, Shuicheng Yan, Wangmeng Zuo
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 teacher ViT meta-training phase (f_g) (ablation) | SUN (NesT) · miniImageNet 5-way 1-shot | accuracy: 49.23 -> 62.40 | Improved |
| add teacher ViT meta-training phase (f_g) (ablation) | SUN (NesT) · miniImageNet 5-way 5-shot | accuracy: 66.57 -> 79.45 | Improved |
| add location-specific supervision (ablation) | SUN (NesT) · miniImageNet 5-way 1-shot | accuracy: 62.40 -> 63.06 | Improved |
| add spatial-consistent augmentation (SCA) (ablation) | SUN (NesT) · miniImageNet 5-way 1-shot | accuracy: 63.06 -> 64.50 | Improved |
| add background filtration (BGF) (ablation) | SUN (NesT) · miniImageNet 5-way 1-shot | accuracy: 64.50 -> 64.84 | Improved |
| add meta-tuning phase (ablation) | SUN (NesT) · miniImageNet 5-way 1-shot | accuracy: 64.84 -> 66.54 | Improved |
| add meta-tuning phase (ablation) | SUN (NesT) · miniImageNet 5-way 5-shot | accuracy: 80.96 -> 82.09 | Improved |
| raise teacher drop path rate from 0.1 to 0.5 (ablation) | teacher ViT · miniImageNet 5-way 1-shot | accuracy: 60.10 -> 62.40 | Improved |
| add CNN branch (inductive bias) to NesT | Meta-Baseline (NesT) · miniImageNet 5-way 1-shot | accuracy: 54.57 -> 57.91 | Improved |
| add CNN distillation to NesT | Meta-Baseline (NesT) · miniImageNet 5-way 1-shot | accuracy: 54.57 -> 55.79 | Improved |
| use local-attention ViT (Swin) instead of global attention (LeViT) | Meta-Baseline · miniImageNet 5-way 1-shot | accuracy: 38.89 -> 54.63 | Improved |
| replace Meta-Baseline with SUN on ViTs | LV-ViT · miniImageNet 5-way 1-shot | accuracy: 43.08 -> 59.00 | Improved |
Closest papers by meaning
- Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning (ICML 2025 · similarity 0.88)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.87)
- One Meta-tuned Transformer is What You Need for Few-shot Learning (ICML 2024 · similarity 0.87)
- A Theoretical Analysis of Self-Supervised Learning for Vision Transformers (ICLR 2025 · similarity 0.87)
- A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.87)
- Vision Transformers Need Registers (ICLR 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.
