TypeError: uint8 类型的 Object 不是 JSON 可序列化的

Question

I'm having some problems with dumping a dictionary to a json file. The dictionary is made up of color tuples/values from an image, and is generated like so:

import cv2, json

img = cv2.imread("bg.png")
HEIGHT, WIDTH = img.shape[:2]

colors = set({})
col_dict = {}
index = 0

for h in range(HEIGHT):
    for w in range(WIDTH):
        if tuple(img[h, w]) in colors:
            continue
        else:
            colors.add(tuple(img[h, w]))
            col_dict[str(index)] = list(img[h, w])
            index += 1

The following loop will extract all the different types of color values in an image and stores them in the dictionary, using the index variable as the key. The colors set variable is used in another part of my program, so it can be ignored.

The loop successfully creates a dictionary, but when executing the following code:

with open("file.json", "w") as infile:
    json.dump(col_dict, infile, indent=2)

The following error is produced:

TypeError: Object of type uint8 is not JSON serializable

The code works when I define a dictionary and it's values manually, like so:

colors = {
    "0" : (109, 87, 34),
    "1" : (209, 87, 34),
    "2" : (85, 98, 34), ...}

with open("file.json", "w") as infile:
    json.dump(colors, infile)

I have no idea why it isn't working now, it still works for other programs that do the same thing. Thanks in advance.

Answer 1

I've found the problem with the code. CV2 image objects are numpy.ndarrays , and each pixel color value is a numpy.uint8 data type. Since json files do not understand numpy uint8 integers, it produced the error TypeError: Object of type uint8 is not JSON serializable .

Here is the solution to my code:

import cv2, json

img = cv2.imread("bg.png")
HEIGHT, WIDTH = img.shape[:2]

colors = set({})
col_dict = {}
index = 0

for h in range(HEIGHT):
    for w in range(WIDTH):
        if tuple(img[h, w]) in colors:
            continue
        else:
            value = (int(img[h, w][0]), int(img[h, w][1]), int(img[h, w][2]))  # This converts each BGR value from uint8 to int.
            colors.add(tuple(img[h, w]))
            col_dict[str(index)] = value
            index += 1

with open("file.json", "w") as infile:
    json.dump(col_dict, infile, indent=2)

All I needed to do was convert the numpy uint8 object to a python integer.