Analysis of scores, datasets, and models in visual saliency modeling • Ali Borji, Hamed R. Tavakoli, Dicky N. Sihite, and Laurent Itti,
Visual Saliency • Why important? • Current status • Methods: numerous / 8 categories (Borji and Itti, PAMI, 2012) • Databases: • Measures: • scan-path analysis • correlation based measures • ROC analysis How good my method works?
Benchmarks • Judd et al. http://people.csail.mit.edu/tjudd/SaliencyBenchmark/ • Borji and Itti https://sites.google.com/site/saliencyevaluation/ • Yet another benchmark!!!?
Toronto MIT Le Meur Dataset Challenge • Dataset bias : • Center-Bias (CB), • Border effect • Metrics are affected by these phenomena.
Tricking the metric Solution ? • sAUC • Best smoothing factor • More than one metric
Features Low level High level people car intensity color symmetry orientation signs depth text size The Feature Crises Does it capture any semantic scene property or affective stimuli? Challenge of performance on stimulus categories & affective stimuli
The Benchmark Image categories and affective data vs 0.64(non-emotional)
aAdDbBcCaA aAcCaA aAcCbBcCaAaA …. The Benchmarkpredicting scanpath aAbBcCaA aA dD bB cC bBbBcC matching score
Lessons learned • We recommend using shuffled AUC score for model evaluation. • Stimuli affects the performance . • Combination of saliency and eye movement statistics can be used in category recognition. • There seems the gap between models and IO is small (though statistically significant). It somehow alerts the need for new dataset. • The challenge of task decoding using eye statistics is open yet. • Saliency evaluation scores can still be introduced