Training with dense depth

[AlfieBrash]

Is it possible to train the model with dense depth annotations? In the form of aligned greyscale png data, which can be converted if needed. (I am experimenting with using synthetic data for training.)
Kind regards.