Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
On the Importance of Distractors for Few-Shot Classification
Rajshekhar Das, Yu-Xiong Wang, José M. F. Moura
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 contrastive finetuning (ConFT) on task samples and distractors to CE-trained prior | CE-trained ResNet10 · Cars 5-shot | accuracy: 51.41 -> 61.53 | Improved |
| add ConFT | CE-trained ResNet10 · CUB 5-shot | accuracy: 62.80 -> 70.53 | Improved |
| add multi-task cross-entropy loss to ConFT | ConFT · CUB 5-shot | accuracy: 70.53 -> 74.45 | Improved |
| add multi-task cross-entropy loss to ConFT | ConFT · Cars 1-shot | accuracy: 39.11 -> 37.36 | Hurt |
| add distractor negatives to task negatives (ablation) | ConFT · Cars 5-shot | accuracy: 59.00 -> 60.58 | Improved |
| add task negatives alongside distractor negatives (ablation) | ConFT · Cars 5-shot | accuracy: 57.42 -> 60.58 | Improved |
| add relative weighting of negatives (ablation) | ConFT · Cars 5-shot | accuracy: 60.58 -> 61.53 | Improved |
| use stochastic pair construction rather than all task positives (ablation) | ConFT · Cars 5-shot | accuracy: 61.01 -> 61.53 | Improved |
| enlarge distractor domain from 512 to 38400 images (ablation) | ConFT · Cars 5-shot | accuracy: 61.16 -> 61.53 | No clear effect |
| use asymmetric rather than standard pair construction (ablation) | ConFT · Cars 1-shot | accuracy: 37.09 -> 39.11 | Improved |
| add augmentation to task-specific samples (ablation) | ConFT · Cars 5-shot | accuracy: 58.64 -> 61.53 | Improved |
| add ConFT finetuning to SimCLR prior | SimCLR Conv4 · miniImageNet 5-shot unsupervised | accuracy: 56.57 -> 60.02 | Improved |
Closest papers by meaning
- Finding Task-Relevant Features for Few-Shot Learning by Category Traversal (CVPR 2019 · similarity 0.91)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.90)
- A Closer Look at Few-shot Classification (ICLR 2019 · similarity 0.89)
- Rethinking Generalization in Few-Shot Classification (NeurIPS 2022 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.89)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.88)
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.
