add support to memmap to take in a list of tiff files with compatible shapes

[kushalkolar]

It might be useful to add support to memmap to take in a list of tif files of compatible shapes and return a numpy.memmap with a shape representing concatenation along one of the axes. I could create a wrapper class for my own use cases but I thought I'd post here because this type of functionality might be useful to other users. I could help implement it (no timeline though) if you think it's worth having.

[cgohlke]

This should already be possible with TiffSequence(...).asarray(out='memmap') but is not available through the memmap function.

[kushalkolar]

Thanks for the fast response! I just tried this and there is a noticeable memory footprint where it seems to be opening each file in RAM as its iterating through the sequence. By creating a wrapper class I was thinking of mapping slices passed to the wrapper instance __getitem__ to the corresponding memmap from a list of memmaps

[kushalkolar]

(does this functionality already exists or is of interest to tifffile)

[cgohlke]

It is currently not possible to pass through an out argument from FileSequence.asarray to the imread function. I'll add it to the next version. Maybe an option can be added to decode directly to the final memory mapped array if the imread function is known to support it...
Anyway, for most cases it would be better to map the segments in each file of the sequence as a zarr store. This is currently only possible for multi-file OME-TIFF or via generating fsspec reference files. A general solution would require support for zarr sharding...

[cgohlke]

Should be fixed in version 2023.1.23.