Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning Instance and Task-Aware Dynamic Kernels for Few-Shot Learning
Rongkai Ma, Pengfei Fang, Gil Avraham, Yan Zuo, Tianyu Zhu, Tom Drummond, Mehrtash Harandi
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 task-specific kernel to support features (ablation) | ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 79.46 -> 81.56 | Improved |
| also apply the task-specific kernel to query features (ablation) | ProtoNet + task kernel · mini-ImageNet 5-way 5-shot | accuracy: 81.56 -> 82.51 | Improved |
| add instance kernel (ablation) | ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 79.46 -> 80.81 | Improved |
| add instance kernel on top of task-specific kernel (ablation) | ProtoNet + task kernel · mini-ImageNet 5-way 5-shot | accuracy: 81.56 -> 81.74 | Improved |
| adapt query features with the task-specific kernel (ablation) | INSTA-ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 81.74 -> 83.13 | Improved |
| use multi-spectral attention instead of global average pooling (ablation) | INSTA-ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 82.07 -> 83.13 | Improved |
| extract an instance kernel from the query too (ablation) | INSTA-ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 83.13 -> 82.24 | No clear effect |
| share the dynamic kernel generator between task and instance kernels (ablation) | INSTA-ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 81.36 -> 83.13 | Improved |
| add residual connection between adapted and original features (ablation) | INSTA-ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 80.03 -> 83.13 | Improved |
| use 3x3 dynamic kernel instead of 5x5 (ablation) | INSTA-ProtoNet · mini-ImageNet 5-way 5-shot | accuracy: 82.43 -> 83.13 | Improved |
| add INSTA to ProtoNet (ablation) | ProtoNet ResNet-12 · mini-ImageNet 5-way 1-shot | accuracy: 62.29 -> 67.01 | Improved |
| add INSTA to DeepEMD (ablation) | DeepEMD ResNet-12 · tiered-ImageNet 5-way 5-shot | accuracy: 86.79 -> 88.02 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepEMD — DeepEMD: Few-Shot Image Classification With Differentiable Earth Mover's Distance and Structured Classifiers (CVPR 2020) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Dynamic Few-Shot Visual Learning Without Forgetting (CVPR 2018 · similarity 0.91)
- Dynamic Convolution: Attention Over Convolution Kernels (CVPR 2020 · similarity 0.89)
- Dense Classification and Implanting for Few-Shot Learning (CVPR 2019 · 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)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.89)
- Looking Wider for Better Adaptive Representation in Few-Shot Learning (AAAI 2021 · 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.
