Text this: CNN-SVO: improving the mapping in semi-direct visual odometry using single-image depth prediction