AI Paper Atlas

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

TPAMI 2018 · Label-efficient and robust learning › Recognition and long-tail · approach: Feature attribution methods · 920 citations (OpenAlex)

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 changedWhereMeasuredEffect
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300KL: 0.00 -> 8.18Hurt
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300IG: 2.50 -> -10.65Hurt
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300SIM: 1.00 -> 0.45Hurt
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300AUC: 0.92 -> 0.67Hurt
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300CC: 1.00 -> 0.50Hurt
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300NSS: 3.29 -> 1.57Hurt
remove 75% of salient pixels from ground truth (false negatives) (ablation)ground-truth fixation map · MIT300EMD: 0.00 -> 1.09Hurt
use permutation control instead of uniform chance modelimage-independent baseline · MIT300KL: 2.09 -> 6.12Hurt
use permutation control instead of uniform chance modelimage-independent baseline · MIT300CC: 0.00 -> 0.20Improved
use center prior instead of single observersaliency baseline · MIT300KL: 6.19 -> 1.24Improved
use single observer instead of center priorsaliency baseline · MIT300NSS: 0.92 -> 1.65Improved
optimize models for center bias, blur and scalesaliency models · MIT300metric rank correlationImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy