This paper addresses the challenges in fixation prediction and salient object segmentation by identifying and addressing the *dataset design bias* in existing benchmarks. The authors propose a new dataset that simultaneously provides both fixation and salient object segmentation ground-truth, aiming to bridge the gap between these two tasks. They find that the widely used salient object segmentation dataset is heavily biased, leading to poor generalization of top-performing algorithms on more realistic images. The paper also highlights the strong correlation between fixations and salient objects, which is leveraged to propose a novel salient object segmentation method. This method combines existing fixation-based saliency models with segmentation techniques, achieving significant improvements over state-of-the-art algorithms on three major datasets. The authors conclude that their dataset and method provide new insights into both fixation prediction and salient object segmentation.This paper addresses the challenges in fixation prediction and salient object segmentation by identifying and addressing the *dataset design bias* in existing benchmarks. The authors propose a new dataset that simultaneously provides both fixation and salient object segmentation ground-truth, aiming to bridge the gap between these two tasks. They find that the widely used salient object segmentation dataset is heavily biased, leading to poor generalization of top-performing algorithms on more realistic images. The paper also highlights the strong correlation between fixations and salient objects, which is leveraged to propose a novel salient object segmentation method. This method combines existing fixation-based saliency models with segmentation techniques, achieving significant improvements over state-of-the-art algorithms on three major datasets. The authors conclude that their dataset and method provide new insights into both fixation prediction and salient object segmentation.