Python image_list to np.array - numpy

Question

I used python's list to add multiple numpy.array images read by opencv:
[array([[[167, 145, 121],
[164, 142, 118],
[167, 145, 121],
...,
[248, 243, 214],
[246, 242, 213],
[249, 245, 216]],
[[172, 150, 126],
[168, 146, 122],
[163, 141, 117],
...,
[249, 244, 214],
[246, 242, 213],
[248, 244, 215]],
...,]
I want to turn the outermost list into a numpy array, that is, a 4-axis tensor np.array:
array([[[[167, 145, 121],
[164, 142, 118],
[167, 145, 121],
...,
[248, 243, 214],
[246, 242, 213],
[249, 245, 216]],
[[168, 146, 122],
[164, 142, 118],
[164, 142, 118],
...,
[248, 243, 214],
[246, 242, 213],
[249, 245, 216]],
[[172, 150, 126],
[168, 146, 122],
[163, 141, 117],
...,
[249, 244, 214],
[246, 242, 213],
[248, 244, 215]],
...,]
However, if I use np.array(mylist) directly, it becomes:
array([array([[[167, 145, 121],
[164, 142, 118],
[167, 145, 121],
...,
[248, 243, 214],
[246, 242, 213],
[249, 245, 216]],
[[168, 146, 122],
[164, 142, 118],
[164, 142, 118],
...,
[248, 243, 214],
[246, 242, 213],
[249, 245, 216]],
...,
[249, 244, 214],
[246, 242, 213],
[248, 244, 215]],
....]
Is there a way to convert this?

Answer 1

Does all the images have the same shape (width, heigh and number of channels)? If so, doing a np.array(mylist) should have worked just fine. For example, here I created 10 random images:
my_list = [np.random.randint(0, 255, size=(1920, 1080, 3), dtype=np.uint8) for i in range(10)]
converted = np.array(my_list)
Which results in what you expects:
array([[[[213, 60, 51],
[229, 125, 207],
[104, 139, 243],
...,
[166, 219, 32],
[116, 27, 108],
[ 99, 79, 21]],
[[176, 141, 170],
[107, 131, 83],
[ 23, 210, 126],
...,
[147, 41, 167],
[203, 118, 86],
[175, 5, 88]]]], dtype=uint8)
Now, if there are images with different shapes, you need to manually define the resulting shape. Otherwise it will fail and give you a warning (VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes is deprecated.)
For instance, I created a random list of images in different sizes and selected the biggest dimension, padding the results with zeros.
my_list = [np.random.randint(0, 255, size=(1920-i, 1080-i, 3), dtype=np.uint8) for i in range(10)]
largest_shape = np.max(np.array([m.shape for m in my_list]), axis=0)
result = np.zeros([len(my_list)]+largest_shape.tolist())
for i, m in enumerate(my_list):
result[i, :m.shape[0], :m.shape[1], :m.shape[2]] = m