Research map / Architectures and efficiency / CNN design
Towards Better Explanations of Class Activation Mapping
Hyungsik Jung, Youngrock Oh
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 |
|---|---|---|---|
| increase number of orderings from 1 to 100 (ablation) | SHAP-CAM · ImageNet | Increase in Confidence (%): 25.9 -> 26.4 | Improved |
| increase number of orderings from 1 to 100 (ablation) | SHAP-CAM · VOC | Increase in Confidence (%): 37.4 -> 43.6 | Improved |
| increase number of orderings from 1 to 100 (ablation) | SHAP-CAM · COCO | Average Drop (%): 23.98 -> 18.27 | Improved |
| increase number of orderings from 1 to 100 (ablation) | SHAP-CAM · ImageNet | Average Drop in Deletion (%): 32.64 -> 33.03 | Improved |
| use SHAP-based coefficients (LIFT-CAM) instead of Grad-CAM | VGG16 CAM · VOC | Increase in Confidence (%): 32.7 -> 38.7 | Improved |
| use LIFT-CAM instead of Ablation-CAM coefficients | VGG16 CAM · COCO | Increase in Confidence (%): 35.0 -> 39.3 | Improved |
| use LIFT-CAM instead of Ablation-CAM coefficients | VGG16 CAM · COCO | Average Drop in Deletion (%): 26.75 -> 26.34 | Hurt |
| use LIFT-CAM instead of Ablation-CAM coefficients | VGG16 CAM · ImageNet | cosine similarity to SHAP values: 0.972 -> 0.980 | Improved |
| use LIFT-CAM instead of Ablation-CAM coefficients | VGG16 CAM · ImageNet | Insertion AUC: 0.4685 -> 0.4712 | Improved |
| use LIFT-CAM instead of Ablation-CAM coefficients | VGG16 CAM · ImageNet | energy in bounding box (%): 51.87 -> 52.43 | Improved |
| use LIFT-CAM instead of Grad-CAM | basic VQA · VQA v2.0 | Average Drop (%): 16.71 -> 14.14 | Improved |
| use LIFT-CAM instead of Grad-CAM | basic VQA · VQA v2.0 | Increase in Confidence (%): 41.95 -> 43.39 | Improved |
Closest papers by meaning
- Empowering CAM-Based Methods with Capability to Generate Fine-Grained and High-Faithfulness Explanations (AAAI 2024 · similarity 0.89)
- DD-CAM: Minimal Sufficient Explanations for Vision Models Using Delta Debugging (NeurIPS 2026 · similarity 0.88)
- Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation (CVPR 2025 · similarity 0.88)
- CAPE: CAM as a Probabilistic Ensemble for Enhanced DNN Interpretation (CVPR 2024 · similarity 0.87)
- Keep the Faith: Faithful Explanations in Convolutional Neural Networks for Case-Based Reasoning (AAAI 2024 · similarity 0.86)
- FastCAV: Efficient Computation of Concept Activation Vectors for Explaining Deep Neural Networks (ICML 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.
