Relevance of a Feed-Forward Model of Visual Attention for Goal-Oriented and Free-Viewing Tasks