Research map / Label-efficient and robust learning / Recognition and long-tail
Sanity Checks for Saliency Metrics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 random RGB perturbation instead of mean perturbation | Faithfulness reliability assessment · CIFAR-10 test | Krippendorff's alpha: 0.18 (0.17---0.20) -> 0.10 (0.09---0.12) | Hurt |
| Use random RGB perturbation instead of mean perturbation | AOPC_MoRF reliability assessment, L=100 · CIFAR-10 test | Krippendorff's alpha: 0.48 (0.46---0.50) -> 0.22 (0.21---0.24) | Hurt |
| Use random RGB perturbation instead of mean perturbation | AOPC_LeRF reliability assessment, L=20 · CIFAR-10 test | Krippendorff's alpha: 0.05 (0.04---0.06) -> 0.13 (0.11---0.15) | Helped |
| Exclude edge detection from method rankings | AOPC_MoRF reliability assessment, L=100, mean perturbation · CIFAR-10 test | Krippendorff's alpha: 0.48 (0.46---0.50) -> 0.37 (0.35---0.39) | Hurt |
| Compare Faithfulness with AOPC_MoRF | Sensitivity saliency maps · CIFAR-10 test | Spearman correlation: 0.34 | No clear effect |
| Compare Faithfulness with AOPC_LeRF | Sensitivity saliency maps · CIFAR-10 test | Spearman correlation: -0.11 | Mixed |
| Use mean perturbation rather than random RGB perturbation | AOPC_MoRF, L=100 · CIFAR-10 test | Krippendorff's alpha: 0.22 (0.21---0.24) -> 0.48 (0.46---0.50) | Helped |
Closest papers by meaning
- Sanity Checks for Saliency Maps (NeurIPS 2018 · similarity 0.90)
- New Definitions and Evaluations for Saliency Methods: Staying Intrinsic, Complete and Sound (NeurIPS 2022 · similarity 0.88)
- Where do Models go Wrong? Parameter-Space Saliency Maps for Explainability (NeurIPS 2022 · similarity 0.87)
- On the Faithfulness of Vision Transformer Explanations (CVPR 2024 · similarity 0.85)
- Advancing Saliency Ranking with Human Fixations: Dataset Models and Benchmarks (CVPR 2024 · similarity 0.84)
- Explainable Saliency: Articulating Reasoning with Contextual Prioritization (CVPR 2025 · similarity 0.83)
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.