Research map / Probabilistic and scientific modeling / Causal discovery
Counterfactual-based Saliency Map: Towards Visual Contrastive Explanations for Neural Networks
Xue Wang, Zhibo Wang, Haiqin Weng, Hengchang Guo, Zhifei Zhang, Lu Jin, Tao Wei, Kui Ren
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 |
|---|---|---|---|
| perturb target class while constraining non-target logits (CCE counterfactuals) instead of EP, PPE, PGD or Gaussian noise | CCE · ImageNet 2012 (ResNet-50) | Class Deviation Score: 4.426 (EP) -> 5.959 | Improved |
| use CCE saliency instead of best baselines in insertion-deletion tests | saliency explanation · ImageNet-compatible (ResNet-50) | Insertion-Deletion difference: 39.95 -> 40.06 | Improved |
| CCE versus propagation-based GBP on deletion metric | CCE · ImageNet-compatible (ResNet-50) | Deletion: 10.72 (GBP) vs 17.96 | Hurt |
| mix positive and negative saliency maps (alpha=0.7) instead of positive only (ablation) | CCE · ImageNet-compatible (ResNet-50) | Insertion-Deletion difference: 39.33 -> 40.06 | Improved |
| mix positive and negative saliency maps instead of positive only (ablation) | CCE · ImageNet-compatible (VGG-19) | Insertion-Deletion difference: 40.07 -> 40.99 | Improved |
| present CCE explanations instead of GradCAM to users | explanation interface · User study (100 participants) | mean confidence rating: 4.022 -> 4.343 | Improved |
| run CCE explanation with sampled counterfactuals and noise instead of one-shot CE | CCE · ImageNet-compatible | running time: CCE takes about five times CE (0.93 s CE, 4.17 s CCE) | Hurt |
| randomize model parameters cascade-wise (sanity check) | CCE · ImageNet-compatible (VGG-19) | SSIM of explanations | Improved |
Closest papers by meaning
- Explainable Saliency: Articulating Reasoning with Contextual Prioritization (CVPR 2025 · similarity 0.89)
- Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition (CVPR 2025 · similarity 0.89)
- Revealing the Gap in Human and VLM Scene Perception through Counterfactual Semantic Saliency (NeurIPS 2026 · similarity 0.88)
- V-CECE: Visual Counterfactual Explanations via Conceptual Edits (NeurIPS 2025 · similarity 0.88)
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.88)
- Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward (CVPR 2026 · 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.
