Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Support and Trivial Prototypes for Interpretable Image Classification
Chong Wang, Yuyuan Liu, Yuanhong Chen, Fengbei Liu, Yu Tian, Davis McCarthy, Helen Frazer, Gustavo Carneiro
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 closeness loss to learn support prototypes near the classification boundary (ablation) | ProtoPNet (ResNet50) · CUB (full) | accuracy (%): 86.5 -> 87.5 | Improved |
| Add discrimination loss to trivial prototypes (ablation) | ProtoPNet (ResNet50) · CUB (full) | accuracy (%): 86.5 -> 87.2 | Improved |
| Add closeness loss (ablation) | ProtoPNet (ResNet34) · Stanford Dogs (full) | accuracy (%): 80.9 -> 83.0 | Improved |
| Combine support and trivial branches (ST-ProtoPNet) instead of support branch alone (ablation) | ST-ProtoPNet (VGG19) · CUB-200-2011 (cropped) | accuracy (%): 81.8 -> 83.2 | Improved |
| Use two different prototype types instead of ensembling two same-type branches (support ensemble) (ablation) | ST-ProtoPNet (VGG19) · CUB-200-2011 (cropped) | accuracy (%): 82.4 -> 83.2 | Improved |
| Use support prototypes instead of trivial prototypes (Trivial ProtoPNet vs Support ProtoPNet) (ablation) | ProtoPNet (VGG19) · CUB-200-2011 (cropped) | accuracy (%): 81.2 -> 81.8 | Improved |
| Add support prototypes to improve interpretability | ProtoPNet (VGG19) · CUB (full) content heatmap | CH (%): 48.66 -> 66.43 (ST-ProtoPNet) | Improved |
| Combine support and trivial prototypes versus support only on IoU (ablation) | ST-ProtoPNet (VGG19) · CUB (full) | IoU (%): 42.04 -> 41.05 | Hurt |
| Use ST-ProtoPNet versus non-interpretable baseline on Cars with ResNet backbones | ST-ProtoPNet (ResNet34) · Stanford Cars (cropped) | accuracy (%): 92.6 -> 91.4 | Hurt |
| Use 40 prototypes per class rather than 10 for ST-ProtoPNet (ablation) | ST-ProtoPNet (VGG16) · CUB (full) | accuracy (%): 76.8 -> 81.0 | Improved |
Closest papers by meaning
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · similarity 0.91)
- MCPNet: An Interpretable Classifier via Multi-Level Concept Prototypes (CVPR 2024 · similarity 0.88)
- ProtoArgNet: Interpretable Image Classification with Super-Prototypes and Argumentation (AAAI 2025 · similarity 0.88)
- Improving Prototypical Visual Explanations with Reward Reweighing, Reselection, and Retraining (ICML 2024 · similarity 0.88)
- Rashomon Sets for Prototypical-Part Networks: Editing Interpretable Models in Real-Time (CVPR 2025 · similarity 0.87)
- A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations (AAAI 2025 · similarity 0.87)
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.
