Research map / Label-efficient and robust learning / Recognition and long-tail
Neural Prototype Trees for Interpretable Fine-Grained Image Recognition
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 |
|---|---|---|---|
| Ensemble 5 ProtoTree models rather than use one ProtoTree | ProtoTree · CUB-200-2011 | Top-1 accuracy: $82.2$ -> $\bm{87.2}$ | Helped |
| Ensemble 5 ProtoTree models rather than use one ProtoTree | ProtoTree · Stanford Cars | Top-1 accuracy: $86.6$ -> $\bm{91.5}$ | Helped |
| Replace ProtoPNet ensemble (3) with ProtoTree ensemble (3) | CUB-200-2011 ensembles · CUB-200-2011 | Top-1 accuracy: $84.8$ -> $86.6$ | Helped |
| Replace ProtoPNet ensemble (3) with ProtoTree ensemble (3) | Stanford Cars ensembles · Stanford Cars | Top-1 accuracy: $91.4$ -> $90.3$ | Hurt |
| Prune leaves with nearly uniform class distributions (ablation) | ProtoTree · CUB-200-2011 | Accuracy: $82.206 \pm 0.723$ -> $82.192 \pm 0.723$ | No clear effect |
| Prune leaves with nearly uniform class distributions (ablation) | ProtoTree · Stanford Cars | Accuracy: $86.584 \pm 0.250$ -> $86.576 \pm 0.245$ | No clear effect |
| After pruning, replace prototypes with nearest latent training patches (ablation) | ProtoTree · CUB-200-2011 | Accuracy: $82.192 \pm 0.723$ -> $82.199 \pm 0.726$ | No clear effect |
| After pruning, replace prototypes with nearest latent training patches (ablation) | ProtoTree · Stanford Cars | Accuracy: $86.576 \pm 0.245$ -> $86.576 \pm 0.245$ | No clear effect |
| Use maximum-path-probability deterministic routing instead of soft routing (ablation) | ProtoTree · CUB-200-2011 | Accuracy: $82.20 \pm 0.01$ -> $82.19 \pm 0.01$ | No clear effect |
| Use greedy deterministic routing instead of soft routing (ablation) | ProtoTree · CUB-200-2011 | Accuracy: $82.20 \pm 0.01$ -> $82.07 \pm 0.01$ | Hurt |
| Use maximum-path-probability deterministic routing instead of soft routing (ablation) | ProtoTree · CUB-200-2011 | Fidelity: n.a. -> $0.999 \pm 0.001$ | Helped |
| Use greedy deterministic routing instead of soft routing (ablation) | ProtoTree · CUB-200-2011 | Fidelity: n.a. -> $0.987 \pm 0.002$ | Helped |
Closest papers by meaning
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · similarity 0.88)
- ProtoArgNet: Interpretable Image Classification with Super-Prototypes and Argumentation (AAAI 2025 · similarity 0.87)
- Interpretable Image Classification with Adaptive Prototype-based Vision Transformers (NeurIPS 2024 · similarity 0.87)
- ViTree: Single-Path Neural Tree for Step-Wise Interpretable Fine-Grained Visual Categorization (AAAI 2024 · similarity 0.87)
- Improving Prototypical Visual Explanations with Reward Reweighing, Reselection, and Retraining (ICML 2024 · similarity 0.85)
- A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations (AAAI 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.