Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Image Deformation Meta-Networks for One-Shot Learning
Check 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 |
|---|---|---|---|
| add learned image deformation augmentation | prototype-based one-shot learning · ImageNet 1K Challenge novel classes, 311-way 5-shot (ResNet-10) | Top-1 accuracy (%): 33.8 -> 39.3 | Helped |
| add learned image deformation augmentation | one-shot learning · miniImageNet 5-way 1-shot | Top-1 accuracy (%): 58.7 -> 59.14±0.86 | Helped |
| add cross-entropy loss for the embedding sub-network (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 21.3 -> 23.1 | Helped |
| add prototype loss for the deformation sub-network (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 15.3 -> 23.1 | Helped |
| rank gallery candidates by predicted class probability (ablation) | IDeMe-Net gallery selection · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 17.0 -> 23.1 | Helped |
| generate deformed images at meta-testing (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 17.0 -> 23.1 | Helped |
| use learned deformation weights instead of random weights (ablation) | IDeMe-Net deformation network · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 15.9 -> 23.1 | Helped |
| use an additional gallery rather than sampling gallery images from the support set (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 17.5 -> 23.1 | Helped |
| learn deformations rather than use gallery images directly (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 15.7 -> 23.1 | Helped |
| use patch-level fusion instead of image-level fusion (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 16.2 -> 23.1 | Helped |
| use patch-level fusion instead of pixel-level fusion (ablation) | IDeMe-Net · ImageNet 1K Challenge novel classes, 311-way 1-shot | Top-1 accuracy (%): 17.3 -> 23.1 | Helped |
| increase the number of synthesized images from 0 to 8 (ablation) | IDeMe-Net · ImageNet 1K Challenge 5-shot | Top-5 accuracy (%) | Helped |
Closest papers by meaning
- Feature Deformation Meta-Networks in Image Captioning of Novel Objects (AAAI 2020 · similarity 0.90)
- A Dual Attention Network with Semantic Embedding for Few-Shot Learning (AAAI 2019 · similarity 0.87)
- Deformable Kernels: Adapting Effective Receptive Fields for Object Deformation (ICLR 2020 · similarity 0.87)
- Low-Shot Learning From Imaginary Data (CVPR 2018 · similarity 0.87)
- Deformable ConvNets V2: More Deformable, Better Results (CVPR 2019 · similarity 0.87)
- Towards flexible perception with visual memory (ICML 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.