An Approach for Preparing Groundtruth Data and Evaluating Visual Saliency Models

Evaluation is a key part while proposing a new model. To evaluate models of visual saliency, one needs to compare the model's output with salient locations in an image. This paper proposes an approach to find out the salient locations, i.e., groundtruth for experiments with visual saliency models. It is found that the proposed human hand-eye coordination based technique can be an alternative to costly human pupil-tracking based systems. Moreover, an evaluation metric is also proposed that suits the necessity of the saliency models.