Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Image Block Augmentation for One-Shot Learning
Zitian Chen, Yanwei Fu, Kaiyu Chen, Yu-Gang Jiang
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 |
|---|---|---|---|
| full Self-Jig vs SVM baseline | SVM one-shot classifier · miniImageNet 5-way 1-shot | accuracy: 52.73 -> 58.80 | Improved |
| full Self-Jig vs SVM baseline | SVM classifier, ResNet-10 · ImageNet1k novel n=1 | top-1: 15.9 -> 23.8 | Improved |
| add self-training gallery selection (ablation) | Rob+Jigsaw (SVM) · miniImageNet 5-way 1-shot | accuracy: 55.54 -> 58.80 | Improved |
| add jigsaw augmentation (ablation) | Rob+Self-T (SVM) · miniImageNet 5-way 1-shot | accuracy: 51.22 -> 58.80 | Improved |
| add robust feature extractor finetuning (ablation) | Self-T+Jigsaw (SVM) · miniImageNet 5-way 1-shot | accuracy: 49.89 -> 58.80 | Improved |
| use only self-training (ablation) | SVM · miniImageNet 5-way 1-shot | accuracy: 52.73 -> 49.81 | Hurt |
| use only robust extractor (ablation) | SVM · miniImageNet 5-way 1-shot | accuracy: 52.73 -> 52.55 | No clear effect |
| use base-class images as gallery instead of unlabeled novel (ablation) | Self-Jig (SVM) · ImageNet1k novel n=1 | top-1: 23.8/51.9 -> 17.7/39.1 | Hurt |
| increase synthesized images from 0 to 10 (ablation) | Self-Jig · miniImageNet 5-way 1-shot | accuracy: 52.55% -> 58.2% | Improved |
| vary number of replaced blocks (ablation) | Self-Jig · miniImageNet 5-way 1-shot | accuracy | No clear effect |
| more shots per class | Self-Jig gain over baseline · ImageNet1k | top-1 | Hurt |
Closest papers by meaning
- Boosting Few-Shot Visual Learning With Self-Supervision (ICCV 2019 · similarity 0.86)
- What Ails One-Shot Image Segmentation: A Data Perspective (NeurIPS 2021 · similarity 0.84)
- Meta-Reinforced Synthetic Data for One-Shot Fine-Grained Visual Recognition (NeurIPS 2019 · similarity 0.84)
- Rethinking Few-shot Image Classification: A Good Embedding is All You Need? (ECCV 2020 · similarity 0.84)
- Less than Few: Self-Shot Video Instance Segmentation (ECCV 2022 · similarity 0.84)
- LibFewShot: A Comprehensive Library for Few-Shot Learning (TPAMI 2023 · similarity 0.84)
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.
