How to store multiple numpy.ndarray into a csv file?

Question

I want to store multiple PIL.Image that are converted to np.ndarray into pd.DataFrame and then store it into a csv file in order to send it to my mlflow model. How can I load the dataframe just like the dataframe that I saved? The data type will change after saving into .csv format as you can see.

Code:

import numpy as np
import pandas as pd
from PIL import Image

img_0 = np.asarray(Image.open("dataset/image_0.jpg"))
img_1 = np.asarray(Image.open("dataset/image_1.jpg"))
df = pd.DataFrame({
    'images': [img_0, img_1],
})
print(df['images'][0])
print(type(df['images'][0]))
df.to_csv('test.csv', index=False)
df = pd.read_csv('test.csv')
print(df['images'][0])
print(type(df['images'][0]))

Output:

<class 'numpy.ndarray'>
<class 'str'>

And the data in df['images'][0] is changed from

[[[ 38  27  23]
  [ 36  25  21]
  [ 33  22  18]
  ...
  [ 30  24  24]
  [ 32  26  26]
  [ 33  27  27]]

 [[ 39  28  24]
  [ 36  25  21]
  [ 33  22  18]
  ...
  [ 30  24  24]
  [ 30  24  24]
  [ 31  25  25]]

 [[ 39  28  24]
  [ 36  25  21]
  [ 34  23  19]
  ...
  [ 29  23  23]
  [ 28  22  22]
  [ 28  22  22]]

 ...

to

'[[[ 38  27  23]\n  [ 36  25  21]\n  [ 33  22  18]\n  ...\n  [ 30  24  24]\n  [ 32  26  26]\n  [ 33  27  27]]\n\n [[ 39  28  24]\n  [ 36  25  21]\n  [ 33  22  18]\n  ...\n  [ 30  24  24]\n  [ 30  24 ...

And the error that I encounter is

Traceback (most recent call last):
  ...
  File "/home/matin/workspace/CI/computational-intelligence-final-project/image_classification_resnet.py", line 158, in <lambda>
    df['images'] = df['images'].apply(lambda x: Image.fromarray(np.uint8(x)))
ValueError: invalid literal for int() with base 10: '[[[ 38  27  23]\n  [ 36  25  21]\n  [ 33  22  18]\n  ...\n  [ 30  24  24]\n  [ 32  26  26]\n  [ 33  27  27]]\n\n [[ 39  28  24]\n  [ 36  25  21]\n  [ 33  22  18]\n  ...\n  [ 30  24  24]\n  [ 30  24