Saliency prediction: What do humans notice about an image?
Much of the deep learning work that takes place at Insight provides us with new perspectives on how the human brain and senses operate.
Experiments with eye trackers have shown that human beings are remarkably consistent in where they direct their attention when shown an image or video. The regions of an image where humans direct their attention are called salient regions. Kevin McGuinness and his team have been the first to apply deep learning to the problem of visual saliency prediction in images. They have achieved state-of-the-art results in video saliency prediction.
Computational models of human visual attention aim to predict how salient each part of an image or video is. These models have many useful applications, including improving the performance of subsequent stages in an image processing pipeline.