Structure-measure: A New Way to Evaluate Foreground Maps This paper introduces a novel structural similarity measure (Structure-measure) to evaluate non-binary foreground maps in salient object detection. Current evaluation measures like AUC, AP, and Fβ often focus on pixel-wise errors and ignore structural similarities, which are crucial for human visual perception. Structure-measure evaluates both region-aware and object-aware structural similarities between a saliency map (SM) and a ground-truth (GT) map. It is efficient, easy to calculate, and outperforms existing measures on five benchmark datasets. The proposed measure is based on two key characteristics: sharp foreground-background contrast and uniform saliency distribution. It combines region-aware and object-aware structural similarity measures. The region-aware measure evaluates structural similarity by dividing the maps into blocks and calculating similarity between them. The object-aware measure evaluates foreground and background regions separately, focusing on their structural similarity. Experiments show that Structure-measure outperforms existing measures in terms of ranking consistency, robustness to annotation errors, and sensitivity to structural changes. It performs better than AP, AUC, and Fβ in several benchmark datasets. A user study further confirms that Structure-measure aligns better with human judgment in evaluating salient object detection models. The paper also compares ten state-of-the-art saliency models on four datasets, showing that Structure-measure ranks models more accurately. Overall, Structure-measure provides a more reliable and effective way to evaluate foreground maps in salient object detection.Structure-measure: A New Way to Evaluate Foreground Maps This paper introduces a novel structural similarity measure (Structure-measure) to evaluate non-binary foreground maps in salient object detection. Current evaluation measures like AUC, AP, and Fβ often focus on pixel-wise errors and ignore structural similarities, which are crucial for human visual perception. Structure-measure evaluates both region-aware and object-aware structural similarities between a saliency map (SM) and a ground-truth (GT) map. It is efficient, easy to calculate, and outperforms existing measures on five benchmark datasets. The proposed measure is based on two key characteristics: sharp foreground-background contrast and uniform saliency distribution. It combines region-aware and object-aware structural similarity measures. The region-aware measure evaluates structural similarity by dividing the maps into blocks and calculating similarity between them. The object-aware measure evaluates foreground and background regions separately, focusing on their structural similarity. Experiments show that Structure-measure outperforms existing measures in terms of ranking consistency, robustness to annotation errors, and sensitivity to structural changes. It performs better than AP, AUC, and Fβ in several benchmark datasets. A user study further confirms that Structure-measure aligns better with human judgment in evaluating salient object detection models. The paper also compares ten state-of-the-art saliency models on four datasets, showing that Structure-measure ranks models more accurately. Overall, Structure-measure provides a more reliable and effective way to evaluate foreground maps in salient object detection.