Deep W-Net: DNN for Spatial Saliency Prediction in Video Frames

Deep neural networks (DNN) have recently been utilized to recognize saliency in images and movies. Various models have been put forth to anticipate spatial, temporal, and spatiotemporal saliency. This paper proposes a new approach for estimating spatial saliency in video frames. Our suggested approach, which achieves excellent results on various measuring parameters, is inspired by U-net architecture. We tested our model using video frame extracts and achieved satisfactory outcomes. Results on data sets with still imagery are also examined. The proposed model does not utilize any transfer learning approaches during any part of training or testing.