Research map / Architectures and efficiency / Pruning and efficient tuning
EfficientFSL: Enhancing Few-Shot Classification via Query-Only Tuning In Vision Transformers
Wenwen Liao, Hang Ruan, Jianbo Yu, Bing Song, Yuansong Wang, Xiaofeng Yang
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 projection layer in Active Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 51.15 -> 80.13 | Improved |
| add trainable prompt Pi (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 39.36 -> 80.13 | Improved |
| add MLP in Active Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 55.09 -> 80.13 | Improved |
| add attention in Active Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 75.98 -> 80.13 | Improved |
| add Combine Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 75.55 -> 80.13 | Improved |
| add Combine Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 5-shot | accuracy (%): 88.60 -> 88.81 | No clear effect |
| add F_att to Combine Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 5-shot | accuracy (%): 89.71 -> 88.81 (5-shot); 78.31 -> 80.13 (1-shot) | Mixed |
| add SQ Attention Block (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 75.42 -> 80.20 | Improved |
| use conditional weights instead of simple average in aggregation (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 69.58 -> 80.13 | Improved |
| use fixed weights instead of simple average in aggregation (ablation) | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 69.58 -> 69.97 | No clear effect |
| use EfficientFSL instead of LoRA | EfficientFSL (ViT-B/16 21K) · FC100 5-way 1-shot | accuracy (%): 74.19 -> 80.13 | Improved |
| use EfficientFSL instead of full fine-tuning ViT-S | EfficientFSL (ViT-S) · FC100 5-way 1-shot | accuracy (%): 52.93 -> 69.94 | Improved |
Closest papers by meaning
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.90)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · 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)
- Scaling Few-Shot Learning for the Open World (AAAI 2024 · similarity 0.88)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.87)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · 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.
