Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Rectifying the Shortcut Learning of Background for Few-Shot Learning
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 |
|---|---|---|---|
| remove background at evaluation (use foreground crops) (ablation) | CC trained on Ori · miniImageNet subset 5-shot | accuracy: 62.73 -> 66.75 | Helped |
| train on foreground-only instead of original images, evaluate on Ori (ablation) | CC · miniImageNet subset 5-shot Dv-Ori | accuracy: 62.73 -> 60.85 | Hurt |
| fusion sampling of foreground and original images at training (ablation) | PN · miniImageNet subset 5-shot Dv-Ori | accuracy: 52.74 -> 59.23 | Helped |
| add COS foreground-aware training sampling (ablation) | CC · miniImageNet 1-shot Dv-Ori | accuracy: 62.67 -> 64.76 | Helped |
| finetune from contrastive pretrained model instead of scratch (ablation) | CC+COS · miniImageNet 5-shot Dv-Ori | accuracy: 81.18 -> 81.16 | No clear effect |
| add SOC evaluation alone without COS (ablation) | CC+FT · miniImageNet 1-shot Dv-Ori | accuracy: 62.71 -> 64.41 | Mixed |
| add SOC on top of COS (ablation) | CC+FT+COS · miniImageNet 1-shot Dv-Ori | accuracy: 65.05 -> 69.29 | Helped |
| replace COS with RBD saliency foreground extraction in training (ablation) | CC · miniImageNet 1-shot | accuracy: 62.67 -> 63.24 | Helped |
| replace COS with MBD saliency in training (ablation) | CC · miniImageNet 1-shot | accuracy: 62.67 -> 61.50 | Hurt |
| replace SOC with RBD saliency crops at evaluation (ablation) | COS · miniImageNet 1-shot | accuracy: 69.28 -> 67.03 | Hurt |
| replace contrastive Exemplar with supervised CC, eval on FG only (ablation) | miniImageNet training · miniImageNet 5-shot Dv-FG | accuracy: 85.12 -> 82.86 | Hurt |
| vary SOC importance factors alpha, beta toward 0 or 1 (ablation) | SOC · miniImageNet | accuracy | Hurt |
Closest papers by meaning
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.87)
- Domain-RAG: Retrieval-Guided Compositional Image Generation for Cross-Domain Few-Shot Object Detection (NeurIPS 2025 · similarity 0.86)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.85)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.85)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.85)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.85)
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.