Visual Attention-based Self-supervised Absolute Depth Estimation using Geometric Priors in Autonomous Driving.