Depth Estimation from Monocular Images Using Dilated Convolution and Uncertainty Learning.

Depth cues are vital in many challenging computer vision tasks. In this paper, we address the problem of dense depth prediction from a single RGB image. Compared with stereo depth estimation, sensing the depth of a scene from monocular images is much more difficult and ambiguous because the epipolar geometry constraints cannot be exploited. In addition, the value of the scale is often unknown in monocular depth prediction. To facilitate an accurate single-view depth prediction, we introduce dilated convolution to capture multi-scale contextual information and then present a deep convolutional neural network. To improve the robustness of the system, we estimate the uncertainty of noisy data by modelling such uncertainty in a new loss function. The experiment results show that the proposed approach outperforms the previous state-of-the-art methods in depth estimation tasks.