Research map / Label-efficient and robust learning / Recognition and long-tail
What Do Different Evaluation Metrics Tell Us About Saliency Models?
Zoya Bylinskii, Tilke Judd, Aude Oliva, Antonio B. Torralba, Frédo Durand
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 |
|---|---|---|---|
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | KL: 0.00 -> 8.18 | Hurt |
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | IG: 2.50 -> -10.65 | Hurt |
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | SIM: 1.00 -> 0.45 | Hurt |
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | AUC: 0.92 -> 0.67 | Hurt |
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | CC: 1.00 -> 0.50 | Hurt |
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | NSS: 3.29 -> 1.57 | Hurt |
| remove 75% of salient pixels from ground truth (false negatives) (ablation) | ground-truth fixation map · MIT300 | EMD: 0.00 -> 1.09 | Hurt |
| use permutation control instead of uniform chance model | image-independent baseline · MIT300 | KL: 2.09 -> 6.12 | Hurt |
| use permutation control instead of uniform chance model | image-independent baseline · MIT300 | CC: 0.00 -> 0.20 | Improved |
| use center prior instead of single observer | saliency baseline · MIT300 | KL: 6.19 -> 1.24 | Improved |
| use single observer instead of center prior | saliency baseline · MIT300 | NSS: 0.92 -> 1.65 | Improved |
| optimize models for center bias, blur and scale | saliency models · MIT300 | metric rank correlation | Improved |
Closest papers by meaning
- Sanity Checks for Saliency Metrics (AAAI 2020 · similarity 0.90)
- Saliency Benchmarking Made Easy: Separating Models, Maps and Metrics (ECCV 2018 · similarity 0.88)
- Spatially Binned ROC: A Comprehensive Saliency Metric (CVPR 2016 · similarity 0.87)
- Evaluation of Visual Processing Should Be Human-Centered, Not Metric-Centered (NeurIPS 2026 · similarity 0.85)
- Multidimensional Observer Model and Perceptual Dimensions of Human Image Quality Assessment (NeurIPS 2026 · similarity 0.85)
- FoMo: Forking Moment in Generative Trajectory as a Perceptual Distance (NeurIPS 2026 · similarity 0.84)
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.
