Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Task-Aware Part Mining Network for Few-Shot Learning
Jiamin Wu, Tianzhu Zhang, Yongdong Zhang, Feng Wu
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 generic part filters (ablation) | ProtoNet · miniImageNet 5-way 1-shot | accuracy: 60.47 -> 64.11 | Improved |
| add task-aware part filters (ablation) | ProtoNet · miniImageNet 5-way 1-shot | accuracy: 60.47 -> 65.94 | Improved |
| use task-aware part filters instead of generic part filters (ablation) | ProtoNet + part filters · miniImageNet 5-way 1-shot | accuracy: 64.11 -> 65.94 | Improved |
| add diversity loss Ldiv (ablation) | ProtoNet + TAPF · miniImageNet 5-way 5-shot | accuracy: 81.47 -> 82.84 | Improved |
| add diversity loss Ldiv (ablation) | ProtoNet + TAPF · miniImageNet 5-way 1-shot | accuracy: 65.94 -> 66.53 | Improved |
| add adaptive importance generator Ga (ablation) | ProtoNet + TAPF · miniImageNet 5-way 5-shot | accuracy: 81.47 -> 82.39 | Improved |
| add adaptive importance generator Ga (ablation) | ProtoNet + TAPF · miniImageNet 5-way 1-shot | accuracy: 65.94 -> 66.83 | Improved |
| add both Ga and Ldiv (full TPMN) (ablation) | ProtoNet + TAPF · miniImageNet 5-way 5-shot | accuracy: 81.47 -> 83.44 | Improved |
| vary number of local parts Nf (ablation) | TPMN · miniImageNet 5-way 1-shot | accuracy: best at Nf = 15 (1-shot), Nf = 20 (5-shot) | Mixed |
| replace ProtoNet global representation with TPMN local part representation | ProtoNet · miniImageNet 5-way 1-shot | accuracy: +7.17% | Improved |
| replace ProtoNet global representation with TPMN local part representation | ProtoNet · miniImageNet 5-way 5-shot | accuracy: +3.97% | Improved |
| use TPMN instead of FEAT under cross-domain transfer | FEAT · miniImageNet to CUB 5-way 1-shot | accuracy: 50.67 -> 52.83 | 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
- Improving Few-Shot Learning through Multi-task Representation Learning Theory (ECCV 2022 · similarity 0.87)
- Revisiting Unsupervised Meta-Learning via the Characteristics of Few-Shot Tasks (TPAMI 2022 · similarity 0.86)
- Adaptive Task Sampling for Meta-Learning (ECCV 2020 · similarity 0.86)
- TADAM: Task dependent adaptive metric for improved few-shot learning (NeurIPS 2018 · similarity 0.85)
- Task-Specific Gradient Adaptation for Few-Shot One-Class Classification (CVPR 2025 · similarity 0.84)
- Beyond Pixel and Object: Part Feature as Reference for Few-Shot Video Object Segmentation (AAAI 2025 · 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.
