将 ndarray 转换为 Base64

Question

A user uploads an image, if that user doesn't have another image to upload then that image is saved. We then take that image, slice it in two halves. The first half gets saved again. As for the second image, we need to convert it into a Base64 Image. However, for some reason I'm getting this error: ValueError: ndarray is not C-contiguous

img = q.choice_set.all()[0].img
reader = misc.imread(img)
height, width, _ = reader.shape
with_cutoff = width // 2
s1 = reader[:, :with_cutoff]
s2 = reader[:, with_cutoff:]
misc.imsave(settings.MEDIA_ROOT + "/" + img.name, s2)
validated_data["choiceimage"] = base64.b64encode(s2)

When I save this in the database, I get an error. What am I doing wrong? How can I decode the numpy array into base64?

Answer 1

If you take a 2D array and extract the left or right half, it will no longer be contiguous in memory - there will be gaps between the rows.

x = np.arange(6).reshape(2,3)

gives x :

array([[0, 1, 2],
       [3, 4, 5]])

If we extract the central column into y :

y = x[:,1:2]

gives y :

array([[1],
       [4]])

but if we check if it is contiguous in memory:

y.flags['C_CONTIGUOUS']

gives:

False

The solution is to extract the column into a contiguous array:

y = np.ascontiguousarray(x[:,1:2])
y.flags['C_CONTIGUOUS']

gives:

True