Research map / Label-efficient and robust learning / Recognition and long-tail
Looking in the Mirror: A Faithful Counterfactual Explanation Method for Interpreting Deep Image Classification Models
Townim Chowdhury, Vu Minh Hieu Phan, Kewen Liao, Nanyu Dong, Minh-Son To, Anton van den Hengel, Johan W. Verjans, Zhibin Liao
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 |
|---|---|---|---|
| add CAM-guided spatial prior (CSP) (ablation) | Mirror-CFE SSC module · CelebA-HQ | validity (Val.): 0.78 -> 0.94 | Improved |
| add CAM-guided spatial prior (CSP) (ablation) | Mirror-CFE SSC module · CelebA-HQ | FID: 40.12 -> 8.97 | Improved |
| add spatial pattern editor over plain skip connection (ablation) | Mirror-CFE SSC module · CelebA-HQ | validity (Val.): 0.03 -> 0.94 | Improved |
| use alpha 0.2 instead of alpha 0 in triangulation loss (ablation) | Mirror-CFE · F-MNIST (k=1) | validity (Val.): 0.905 -> 0.949 | Improved |
| add triangulation loss (alpha 0.2) (ablation) | Mirror-CFE · F-MNIST (k=1) | denoised validity (D.Val): 0.921 -> 0.960 | Improved |
| add triangulation loss (alpha 0.2) (ablation) | Mirror-CFE · F-MNIST (k=1) | validity (Val.): 0.971 -> 0.949 | Hurt |
| change alpha among 0, 0.2, 0.4, 0.6 (ablation) | Mirror-CFE · F-MNIST (k not 1) | L1: 0.126 -> 0.127 -> 0.128 -> 0.116 | No clear effect |
| train multiclass instead of pairwise (ablation) | Mirror-CFE · MNIST | FID: 3.20 -> 6.94 | Hurt |
| train multiclass instead of pairwise (ablation) | Mirror-CFE · F-MNIST | L1: 0.25 -> 0.20 | Improved |
| move from 1st CFE to reflection point k=1 | Mirror-CFE · B-MNIST | FID: 86.02 -> 132.93 | Hurt |
| move from 1st CFE to reflection point k=1 | Mirror-CFE · F-MNIST | denoised validity (D.Val): 0.99 -> 0.96 | Hurt |
| use Mirror-CFE instead of C3LT | Mirror-CFE (k=1) · MNIST | denoised validity (D.Val): 0.79 -> 0.99 | Improved |
Closest papers by meaning
- Rethinking Visual Counterfactual Explanations Through Region Constraint (ICLR 2025 · similarity 0.90)
- Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition (CVPR 2025 · similarity 0.88)
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.88)
- DiG-IN: Diffusion Guidance for Investigating Networks - Uncovering Classifier Differences Neuron Visualisations and Visual Counterfactual Explanations (CVPR 2024 · similarity 0.87)
- V-CECE: Visual Counterfactual Explanations via Conceptual Edits (NeurIPS 2025 · similarity 0.87)
- LeapFactual: Reliable Visual Counterfactual Explanation Using Conditional Flow Matching (NeurIPS 2025 · 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.
