Add 2d tiff backend

[fm3]

requires tiled tifs with resolution pyramid, as produced by vips tiffsave source.tif color.tif --tile --pyramid --bigtiff --compression none --tile-width 256 --tile-height 256 (make sure you have enough space at /tmp or change your TMPDIR env var) or convert source.tif -define tiff:tile-geometry=256x256 'ptif:color.tif' (note that imagemagick/convert fails for >4GB images, as it does not support resolution pyramid and bigtiff at the same time)

In data/datasets.json, add

    "tiff": {
        "sample_organization": {
            "my_tiff_dataset": {
                "scale": [1,1]
            }
        }
    }

and in data/binary/sample_organization/my_tiff_dataset, place the generated color.tif

TODO:

• [x] read tiles from tifs using numpy.mmap
• [x] handle resolution pyramid
• [x] add caching, add clear-cache-route
• [x] how to discern layer categories? segmentation layers? → layer_name.startswith("segmentation")
• [x] handle multiple layers
• [x] handle dtypes other than uint8
• [x] handle tif filenames other than "0.tif"
• [x] add readable error messages for incompatible tifs
• [x] test with bigtiff (>4GB)

[fm3]

@jstriebel I implemented your feedback and tested with the large tif (had to introduce the dtype adapter for signed data). could you have another look?

[fm3]

Done and done :) I don’t think it makes that much sense to include a tiff dataset by default in the repo, as we don’t have a real-world dataset that we can use and I don’t think we should include too much stuff that users have to remove first when setting this up for real. What do you think?