`LifImage.as_array()` fails for 4D image

[lanery]

Main issue

Trying to load a 4D (X, Y, Λ, t) LIF image as a numpy array failed.

Code that produced the issue

The file in question is 6 GB, so it's too large to upload here, but it was read in via:

from pathlib import Path
from readlif.reader import LifFile

lif_file = LifFile("filepath/to/4d_image.lif")
lif_image = lif_file.get_image_by_name("Lambda_time_series")

Check the number of "planes" in each dimension:

>>> lif_image.dims_n
{1: 512, 2: 512, 9: 60, 4: 99}

Try to get image data as a numpy array:

>>> lif_image.as_array()
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[70], line 1
----> 1 lif_image.as_array()

File ~/Repositories/readlif/readlif/reader.py:484, in LifImage.as_array(self, dims)
    479 # multichannel image -- get planes recursively for each dimension
    480 array = [
    481     [self.get_plane(requested_dims={dim: i}) for i in range(self.dims_n[dim])]
    482     for dim in dims
    483 ]
--> 484 return np.array(array).squeeze()

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (2,) + inhomogeneous part.

It fails because the as_array() method does not properly iterate over multiple dimensions.

[lanery]

Possible partial fix

Something along the lines of this:

from itertools import product

lif_image = lif_file.get_image_by_name("Lambda_time_series")
dimensions = [i for i in range(3, 10) if i in lif_image.dims_n.keys()]
shape = tuple(lif_image.dims_n.values())

plane_ranges = [range(lif_image.dims_n[dim]) for dim in dimensions]
planes = []
for plane_indices in product(*plane_ranges):
    requested_dims = {dim: i for dim, i in zip(dimensions, plane_indices, strict=True)}
    plane = lif_image.get_plane(requested_dims=requested_dims)
    planes.append(plane)

image = np.array(planes).reshape(shape[::-1])

The above code seems to get the dimensions of the numpy array correct, at least for this particular LIF image:

>>> image.shape
(99, 60, 512, 512)  # (t, Λ, Y, X)

But I'm not sure how well this holds in general without more testing.