Splitting up depth prediction and stabilisation.

[luuude]

You mention in the instructions that we can split up the tasks to save some memory. How would I go about doing that? Save the depth as arrays and write a new inference.py code? Or is it supported with some arguments?