Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Revisiting Local Descriptor Based Image-To-Class Measure for Few-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 |
|---|---|---|---|
| Use learnable deep local descriptors and episodic image-to-class matching instead of image-level feature matching | DN4 on miniImageNet · miniImageNet 5-way 1-shot | accuracy: 50.44% to 51.24% | Helped |
| Use learnable deep local descriptors and episodic image-to-class matching instead of image-level feature matching | DN4 on miniImageNet · miniImageNet 5-way 5-shot | accuracy: 66.53% to 71.02% | Helped |
| Add training data augmentation by random cropping and horizontal flipping (ablation) | DN4 on fine-grained datasets · Stanford Dogs, Stanford Cars, and CUB-200 | accuracy: 63.51 ±0.62 -> 66.33 ±0.66; 88.65 ±0.44 -> 89.60 ±0.44; 74.92 ±0.64 -> 81.90 ±0.60 | Helped |
| Replace image-to-class matching with concatenated global image descriptors and image-to-image matching (ablation) | DN4 · miniImageNet 5-way 1-shot and 5-shot | accuracy: 51.24 ±0.74 -> 37.39 ±0.82; 71.02 ±0.64 -> 50.47 ±0.66 | Hurt |
| Restrict local-descriptor neighbor search to each individual support image instead of the whole support class (ablation) | DN4 · miniImageNet 5-way 5-shot | accuracy: 71.02 ±0.64 -> 69.52 ±0.62 | Hurt |
| Set the number of neighbors k to 1 rather than 3 (ablation) | DN4 · miniImageNet 5-way 5-shot | accuracy: 71.02 -> 71.95 | Helped |
| Set the number of neighbors k to 5 rather than 3 (ablation) | DN4 · miniImageNet 5-way 5-shot | accuracy: 71.02 -> 70.20 | Hurt |
| Set the number of neighbors k to 7 rather than 3 (ablation) | DN4 · miniImageNet 5-way 5-shot | accuracy: 71.02 -> 68.56 | Hurt |
| Replace Conv-64F with ResNet-256F (ablation) | DN4 · miniImageNet 5-way 1-shot and 5-shot | accuracy: 51.24 ±0.74 -> 54.37±0.36%; 71.02 ±0.64 -> 74.44±0.29% | Helped |
| Train with 5 shots instead of matching 1-shot training (ablation) | DN4 · miniImageNet 5-way 1-shot | accuracy: 51.24 -> 53.85 | Helped |
Closest papers by meaning
- DMN4: Few-Shot Learning via Discriminative Mutual Nearest Neighbor Neural Network (AAAI 2022 · similarity 0.92)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · similarity 0.90)
- Rethinking Generalization in Few-Shot Classification (NeurIPS 2022 · similarity 0.89)
- Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference (CVPR 2022 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement (ICLR 2025 · similarity 0.86)
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.