Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Benchmarking Spurious Bias in Few-Shot Image Classifiers
Guangtao Zheng, Wenqian Ye, Aidong Zhang
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 |
|---|---|---|---|
| construct test tasks with FewSTAB attribute-based selection instead of random sampling | ten few-shot classifiers · miniImageNet, tieredImageNet, CUB-200 | wAcc: RENet miniImageNet 5-shot 56.52 -> 36.19 | Hurt |
| add inter-class attribute-based sample selection to intra-class selection (ablation) | FewSTAB · miniImageNet 5-way 5-shot | average accuracy drop (%): 5.13 -> 13.30 | Improved |
| add likelihood-score based query sample selection (ablation) | FewSTAB · miniImageNet 5-way 5-shot | average accuracy drop (%): 13.30 -> 15.05 | Improved |
| switch from 1-shot to 5-shot test tasks | FewSTAB · Spearman rank correlation wAcc-A vs wAcc-R | rank correlation: CUB-200 1.00 -> 0.94 | Mixed |
| increase training shots from 1-shot to 5-shot for meta-learning methods | meta-learning FSC methods · FewSTAB tasks with 1/5/10 test shots | wAcc-A gap | Mixed |
| replace ViT-GPT2 with BLIP as the attribute-detecting VLM (ablation) | FewSTAB · miniImageNet 5-shot | Spearman rank correlation of wAcc-A: 1.00 | No clear effect |
| increase test shots from 1 to 5 in FewSTAB | FewSTAB · miniImageNet, tieredImageNet, CUB-200 | accuracy gap (wAcc-R minus wAcc-A) | Improved |
Closest papers by meaning
- Dataset Bias in Few-Shot Image Recognition (TPAMI 2022 · similarity 0.90)
- Worst Case Matters for Few-Shot Recognition (ECCV 2022 · similarity 0.90)
- On the Importance of Firth Bias Reduction in Few-Shot Classification (ICLR 2022 · similarity 0.90)
- AMU-Tuning: Effective Logit Bias for CLIP-based Few-shot Learning (CVPR 2024 · similarity 0.88)
- Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement (ICLR 2025 · similarity 0.87)
- MetaCoCo: A New Few-Shot Classification Benchmark with Spurious Correlation (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.
