Lighter and Faster Cross-Concatenated Multi-Scale Residual Block Based Network for Visual Saliency Prediction