Research map / Vision-language and multimodal / Human and embodied modeling
ConEx: Human-Interpretable Saliency Maps via Concept-Aware Attribution
Yehonatan Elisha, Oren Barkan, Ziv Weiss Haddad, Noam Koenigstein
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 |
|---|---|---|---|
| build CAVs from concept segments instead of full images (ablation) | ConEx · ImageNet (RN, CN, DN) | VCM: 0.59 -> 0.68 | Improved |
| build CAVs from concept segments instead of full images (ablation) | ConEx · ImageNet (RN, CN, DN) | CCM: 24.68 -> 27.83 | Improved |
| use ConEx centroid-difference CAV construction instead of MCD's CAV construction (ablation) | ConEx · ImageNet RN | INS: 58.16 -> 59.21 | Improved |
| use ConEx CAV construction instead of ACE's (ablation) | ConEx · ImageNet RN | DEL: 9.58 -> 8.34 | Improved |
| use LRP multiplicative fusion instead of GradCAM fusion (ablation) | ConEx · ImageNet RN | INS: 58.68 -> 59.21 | Improved |
| use LRP multiplicative fusion instead of SHAP fusion (ablation) | ConEx · ImageNet RN | INS: 56.99 -> 59.21 | Improved |
| use LRP multiplicative fusion instead of IIA fusion (ablation) | ConEx · ImageNet RN | DEL: 8.65 -> 8.34 | Improved |
| wrap GradCAM with the ConEx concept framework | GradCAM · ImageNet RN | INS: 54.92 -> 58.68 | Improved |
| wrap SHAP with the ConEx concept framework | SHAP · ImageNet RN | INS: 47.54 -> 56.99 | Improved |
| wrap IIA with the ConEx concept framework | IIA · ImageNet RN | DEL: 9.48 -> 8.65 | Improved |
| use the last layer instead of layer 12 -> middle layer L=6 for ViT-B segment embeddings (ablation) | ConEx · ImageNet ViT-B | CINS: 32.40 -> 35.38 | Improved |
| use last layer L instead of L-2 for RN segment embeddings (ablation) | ConEx · ImageNet RN | CCM: 27.56 -> 28.19 | Improved |
Closest papers by meaning
- Explainable Saliency: Articulating Reasoning with Contextual Prioritization (CVPR 2025 · similarity 0.89)
- Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward (CVPR 2026 · similarity 0.89)
- ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features (ICML 2025 · similarity 0.88)
- Probabilistic Conceptual Explainers: Trustworthy Conceptual Explanations for Vision Foundation Models (ICML 2024 · similarity 0.86)
- On the Faithfulness of Vision Transformer Explanations (CVPR 2024 · similarity 0.86)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.86)
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.
