Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Worst Case Matters for Few-Shot Recognition
Minghao Fu, Yun-Hao Cao, Jianxin Wu
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 stability regularization (SR) on top of AC (ablation) | AC fine-tuning, WRN-28-10 · mini-ImageNet 5-way 1-shot | ACC_1: 36.80 -> 40.52 | Improved |
| add stability regularization (SR) on top of AC (ablation) | AC fine-tuning, ResNet-18 · mini-ImageNet 5-way 1-shot | ACC_1: 25.60 -> 32.00 | Improved |
| add ensemble of SR (EnSR) on top of SR (ablation) | AC+SR, WRN-28-10 · mini-ImageNet 5-way 1-shot | ACC_m: 69.38 -> 69.59 | Improved |
| add adaptability calibration alone (fine-tune res5) (ablation) | ResNet-18 baseline · mini-ImageNet 5-way 1-shot | ACC_1: 30.42 -> 25.60 | Hurt |
| add adaptability calibration alone (fine-tune res5) (ablation) | ResNet-18 baseline · mini-ImageNet 5-way 5-shot | ACC_m: 73.61 -> 78.58 | Improved |
| add adaptability calibration alone (ablation) | WRN-28-10 baseline · mini-ImageNet 5-way 1-shot | ACC_m: 67.86 -> 67.36 | Hurt |
| add AC+SR to Baseline pretraining (ablation) | Baseline-pretrained ResNet-18 · mini-ImageNet 5-way 1-shot | ACC_1: 18.42 -> 21.86 | Improved |
| add AC+SR to Baseline pretraining (ablation) | Baseline-pretrained ResNet-18 · mini-ImageNet 5-way 1-shot | ACC_m: 50.37 -> 50.27 | No clear effect |
| add AC+EnSR to Baseline pretraining (ablation) | Baseline-pretrained ResNet-18 · mini-ImageNet 5-way 1-shot | ACC_m: 50.37 -> 54.42 | Improved |
| add AC+EnSR to Baseline++ pretraining (ablation) | Baseline++-pretrained ResNet-18 · mini-ImageNet 5-way 5-shot | ACC_1: 49.08 -> 56.80 | Improved |
| fine-tune more residual groups (res5+res4+res3+res2+1) instead of res5 only (ablation) | AC+EnSR, ResNet-18 · mini-ImageNet 5-way 1-shot | ACC_1: 32.78 -> 25.60 | Hurt |
| fine-tune res5+res4+res3 instead of res5 only (ablation) | AC+EnSR, ResNet-18 · mini-ImageNet 5-way 1-shot | ACC_1: 32.78 -> 22.96 | Hurt |
Closest papers by meaning
- Benchmarking Spurious Bias in Few-Shot Image Classifiers (ECCV 2024 · similarity 0.90)
- Dataset Bias in Few-Shot Image Recognition (TPAMI 2022 · similarity 0.89)
- LibFewShot: A Comprehensive Library for Few-Shot Learning (TPAMI 2023 · similarity 0.88)
- Meta-Learning to Detect Rare Objects (ICCV 2019 · similarity 0.88)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.88)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (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.
