Text this: Joint Soft–Hard Attention for Self-Supervised Monocular Depth Estimation