Performance Assessment Framework for Computational Models of Visual Attention

This paper presents performance framework for computational model of visual attention, a software package, written using python scripting language, developed for the real-time comparison of computational model with human fixations. The performance framework was developed for real-time processing of eye trackers recorded data, analyzing them to generate fixation map, and comparing the fixation map to a saliency model got by running a configured computational model either in bottom-up or top-down mode. The framework is designed such that added modules can be extended for various experiment processing as required by the researcher. The framework encompasses the main connection to eye tracker to collect the raw data that will have observers eye coordinates and duration, it has analysis model to analyze the model and providing methods of visualization like fixation, heatmap and scanpath, it also has a computational model that predicts the fixation on the given image stimulus, finally the platform compares the fixation and saliency map to assess the accuracy of the prediction. All the functions of the framework can be controlled by using the graphical user interfaces.