The document presents Deep Virtual Stereo Odometry (DVSO), a novel monocular visual odometry system that utilizes deep depth prediction to mitigate scale drift and improve the accuracy of 3D reconstruction. The approach trains a depth estimator using stereo camera inputs, which are then applied during inference with monocular cameras for depth initialization in Direct Sparse Odometry. Experimental results demonstrate that DVSO achieves performance comparable to stereo methods while outperforming state-of-the-art monocular visual odometry techniques.