Image.fromarray silently fails with floating-point input

[jakevdp]

When passing a floating-point array to Image.fromarray, the output does not match the input:

from PIL import Image
import numpy as np
x = np.ones((100, 100), dtype='float64')
x[40:60, 40:60] = 0.0
Image.fromarray(x, mode='L')

This happens is because when mode is specified, the array's bytes are passed directly to frombuffer without any type check:

Pillow/PIL/Image.py

Lines 2394 to 2438 in c82f9fe

	def fromarray(obj, mode=None):
	"""
	Creates an image memory from an object exporting the array interface
	(using the buffer protocol).
	If obj is not contiguous, then the tobytes method is called
	and :py:func:`~PIL.Image.frombuffer` is used.
	:param obj: Object with array interface
	:param mode: Mode to use (will be determined from type if None)
	See: :ref:`concept-modes`.
	:returns: An image object.
	.. versionadded:: 1.1.6
	"""
	arr = obj.__array_interface__
	shape = arr['shape']
	ndim = len(shape)
	strides = arr.get('strides', None)
	if mode is None:
	try:
	typekey = (1, 1) + shape[2:], arr['typestr']
	mode, rawmode = _fromarray_typemap[typekey]
	except KeyError:
	# print(typekey)
	raise TypeError("Cannot handle this data type")
	else:
	rawmode = mode
	if mode in ["1", "L", "I", "P", "F"]:
	ndmax = 2
	elif mode == "RGB":
	ndmax = 3
	else:
	ndmax = 4
	if ndim > ndmax:
	raise ValueError("Too many dimensions: %d > %d." % (ndim, ndmax))
	size = shape[1], shape[0]
	if strides is not None:
	if hasattr(obj, 'tobytes'):
	obj = obj.tobytes()
	else:
	obj = obj.tostring()
	return frombuffer(mode, size, obj, "raw", rawmode, 0, 1)

It would be helpful for users if the dtype of the input were checked, and an appropriate warning or error raised.

I'm using Python 3.6.1 and PIL v4.2.1

[wiredfool]

My impression of this parameter (which, tbh, is not well documented) is that it's for overriding the mode that's detected from the dtype. Raising an error where it doesn't match is going to be problematical for the intended use case. On the other hand, since it's assigned directly to the rawmode as well as the mode, there are only a handful of valid modes that will work, generally the ones where the mode and rawmode match in the unpackers table in libImaging/Unpack.c

[radarhere]

#4887 encountered this as well.

import numpy as np
from PIL import Image

array = np.full((32,32,3),255)
Image.fromarray(array, "RGB")

And #5227

import numpy as np
from PIL import Image

vector = np.ones((5, 5, 3)).astype("uint8") * np.array([3])
Image.fromarray(vector, mode="RGB")

And #5465

import numpy as np
from PIL import Image
tgt = np.ones(shape=(2, 2)) # float32
tgt_im = Image.fromarray(tgt)
tgt_im_np = np.asarray(tgt_im)
final_img = Image.fromarray(tgt_im_np, mode='L')

print(np.asarray(final_img))

[radarhere]

If you remove the mode parameter from the original post's fromarray call,

from PIL import Image
import numpy as np
x = np.ones((100, 100), dtype='float64')
x[40:60, 40:60] = 0.0
im = Image.fromarray(x)
im.save("out.tiff")

then you get the expected result - out.tiff.zip
(Here is it converted to PNG since GitHub won't allow TIFFs)

I'm going to suggest resolving this by just improving the documentation in #5849

[maoshi7]

from PIL import Image
import numpy as np

a = np.zeros((200,200))
a[0:2,0:2] = 1 
a[99:101,99:101] = 1
img = Image.fromarray(a,mode='L')
img1 = np.array(img)

I find that a is a matrix with lots of zeros and 8 ones at the right position, but img1 is a wrong matrix. Wrong numbers 240 and 63 appears at strange positions:

There is no warning, I didn`t understand why this change happened until I see this issue.
I would like to know What data types and ranges are required for the different modes of Image.fromarray? I don't see this in the official documentation of pillow, can someone tell me? Did I mislocate the official documentation?
https://pillow-docs-cn.readthedocs.io/zh-cn/latest/reference/Image.html#PIL.Image.fromarray
Thanks!