Research map / Label-efficient and robust learning / Recognition and long-tail
Visual correspondence-based explanations improve AI robustness and human-AI team accuracy
Mohammad Reza Taesiri, Giang Nguyen, Anh Totti Nguyen
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 |
|---|---|---|---|
| Use training-set exemplars with kNN instead of the ResNet-50 classifier | ImageNet classifiers · Adversarial Patch | top-1 accuracy (%): 55.04 -> 59.30 | Improved |
| Rerank kNN candidates using five patch correspondences with EMD | kNN · ImageNet-R | top-1 accuracy (%): 36.18 -> 37.75 | Improved |
| Rerank kNN candidates using five patch correspondences with CHM | kNN · DAmageNet | top-1 accuracy (%): 7.59 -> 8.10 | Improved |
| Use EMD-Corr rather than kNN | ImageNet classifiers · ImageNet | top-1 accuracy (%): 74.77 -> 74.93 | Hurt |
| Use EMD-Corr rather than kNN | CUB with iNaturalist features · CUB | top-1 accuracy (%): 85.46 -> 84.98 | Hurt |
| Replace automatically selected patches with five human-defined bird keypoints (ablation) | CHM-Corr · CUB (iNaturalist features) | top-1 accuracy (%): 83.27 -> 81.54 | Hurt |
| Increase predefined bird keypoints from five to 10 (ablation) | CHM-Corr+ · CUB | top-1 accuracy (%): 81.51 -> 82.34 | Improved |
| Increase predefined bird keypoints from five to 15 (ablation) | CHM-Corr+ · CUB | top-1 accuracy (%): 81.51 -> 82.27 | Improved |
| Use all 49×49 patch pairs instead of five cross-correlation-selected pairs (ablation) | EMD-Corr · CUB (iNaturalist ResNet) | accuracy (%): 84.98 -> 85.42 | Improved |
| Use uniform rather than cross-correlation patch weights (ablation) | EMD-Corr · CUB (iNaturalist ResNet) | accuracy (%): 84.98 -> 79.72 | Hurt |
| Show patch-correspondence boxes in the explanation (ablation) | EMD-Corr human-AI explanations · CUB human study | user accuracy: 64.12 -> 67.64 | Improved |
| Show patch-correspondence boxes in the explanation (ablation) | CHM-Corr human-AI explanations · CUB human study | user accuracy: 65.72 -> 69.72 | Improved |
Closest papers by meaning
- I-CEE: Tailoring Explanations of Image Classification Models to User Expertise (AAAI 2024 · similarity 0.90)
- HIVE: Evaluating the Human Interpretability of Visual Explanations (ECCV 2022 · similarity 0.88)
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.88)
- Interpretable Basis Decomposition for Visual Explanation (ECCV 2018 · similarity 0.88)
- Comparing the Decision-Making Mechanisms by Transformers and CNNs via Explanation Methods (CVPR 2024 · similarity 0.87)
- Human and AI Perceptual Differences in Image Classification Errors (AAAI 2025 · 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.
