PIL 双三次插值法

Question

So, I have an ndarray of the shape: (28, 28, 1) and I want to use bicubic interpolation to upscale it to (33, 33, 1) .
Here's what I got by searching online:

import numpy as np
from PIL import Image as pil_image

...

img = np.array(pil_image.fromarray(img).resize(intermediate_dim, pil_image.BICUBIC))
# print(img.shape) -> (28, 28, 1), print(intermediate_dim) -> (33, 33)

But I keep getting an error:

Traceback (most recent call last):
File "filters.py", line 256, in
visualize_layer(mnist, LAYER_NAME)
File "filters.py", line 232, in visualize_layer
img_loss = _generate_filter_image(input_img, output_layer.output, f)
File "filters.py", line 161, in _generate_filter_image img = np.array(pil_image.fromarray(img).resize(intermediate_dim, pil_image.BICUBIC))
File "BlaBlaBla\\Python37\\site-packages\\PIL\\Image.py", line 2682, in fromarray
raise TypeError("Cannot handle this data type: %s, %s" % typekey)
TypeError: Cannot handle this data type: (1, 1, 1), |u1

After seeing this, I tried using img.astype(np.uint8) but nothing helped.
What am I doing wrong?

PS I am certain that the array contains legal values (integers from [0, 255])

Answer 1

The array you are trying to create a PIL Image from is shaped:

(28,28,1)

and that is not a shape that PIL likes. It likes:

(M,N) for greyscale images
(M,N,3) for RGB images
(M,N,4) for RGBA images

So, you can use the following to select the first channel of that last unused dimension:

img = np.array(pil_image.fromarray(img[...,0]).resize(intermediate_dim, pil_image.BICUBIC))