Hi, and thanks for the code! I'm starting to work with the patch-based memory-mapping approach for handling large datasets (~150GB for me typically). I was hoping to modify the methods to use substantial temporal and spatial downsampling at initialization, but relax this progressively to eventually extract signals at full spatio-temporal resolution. Would a reasonable way to do this be as straightforward as simply resizing the spatial and temporal components after initialization and before refinement? Are there any concerns with this approach to keep in mind? Thanks!