Saliency Map Generation Based on Human Level Performance

Generating precise saliency maps from eye tracker fixation points is a challenging task influenced by environmen-tal factors and the choice of evaluation metrics. This paper presents a novel, sustainable, scale-invariant, and sampling-independent method for converting fixation points into saliency maps. Leveraging the inherent predictability of human behavior, the proposed method ensures the highest compatibility with the chosen evaluation metric. Moreover, it introduces a mechanism to calculate the maximum achievable similarity score for each conversion. In addition, it offers crucial insights for both saliency map evaluation and the training of machine learning systems dedicated to saliency map generation. Experimental results demonstrate the method's efficacy in producing saliency maps that align seamlessly with diverse evaluation metrics, showcasing its adaptability and predictive capabilities. This approach con-tributes not only to the refinement of saliency map generation but also to the broader understanding of the intricacies involved in converting eye tracker data into meaningful ground truths.