Research map / Label-efficient and robust learning / Recognition and long-tail
OPAL: Orthonormal Prototype Alignment Learning for Interpretable Image Classification
Ilán Carretero, Gustavo ANGULO, Rocío Amor, Valery Naranjo
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 channel-wise softmax (ablation) | OPAL (Proj+GMP+Orth) · CUB spatial prototype collapse | SPC: 77.3 -> 1.7 | Improved |
| add channel-wise softmax (ablation) | OPAL (Proj+GMP+Orth) · DOGS spatial prototype collapse | SPC: 93.6 -> 4.9 | Improved |
| add channel-wise softmax (ablation) | OPAL (Proj+GMP+Orth) · PETS | top-1 accuracy (%): 93.2 -> 92.8 | Hurt |
| add channel-wise softmax (ablation) | OPAL (Proj+GMP+Orth) · CUB | top-1 accuracy (%): 82.0 -> 85.6 | Improved |
| add orthonormal anchors (ablation) | Proj-only network · CUB | top-1 accuracy (%): 82.3 -> 85.0 | Improved |
| add orthonormal anchors (ablation) | Proj-only network · CARS | top-1 accuracy (%): 86.3 -> 90.6 | Improved |
| add global max pooling without orthonormal anchors (ablation) | ConvNeXt-Tiny · CARS | top-1 accuracy (%): 87.8 -> 75.7 | Hurt |
| add global max pooling to Proj+CWS+Orth (ablation) | OPAL · CARS | top-1 accuracy (%): 90.4 -> 90.3 | No clear effect |
| add channel-wise softmax only (ablation) | ConvNeXt-Tiny · CARS | top-1 accuracy (%): 87.8 -> 48.9 | Hurt |
| use m=1 prototype per class instead of larger m (ablation) | OPAL · five fine-grained benchmarks (avg) | top-1 accuracy: -1.04 points vs baseline | Hurt |
| use m in {3,5,10} prototypes per class (ablation) | OPAL · five fine-grained benchmarks (avg) | top-1 accuracy: +1.25 points vs baseline | Improved |
| add OPAL to ResNet-50 | ResNet-50 · CARS | top-1 accuracy (%): 71.9 -> 84.3 | 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.
- PIP-Net — PIP-Net: Patch-Based Intuitive Prototypes for Interpretable Image Classification (CVPR 2023) · better on 5 of 5 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
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · similarity 0.87)
- Interpretability Benchmark for Evaluating Spatial Misalignment of Prototypical Parts Explanations (AAAI 2024 · similarity 0.86)
- Interpretable Image Classification with Adaptive Prototype-based Vision Transformers (NeurIPS 2024 · similarity 0.86)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.85)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · similarity 0.85)
- ProtoArgNet: Interpretable Image Classification with Super-Prototypes and Argumentation (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.
