To write a JSON file for a dataset in the COCO format I have an RLE encoded mask with type(counts) == bytes.
To write the dict in a JSON file I use json.dumps(). This works for the strings, integers and floats in the dict, but not for the bytes element.

TypeError: Object of type ndarray is not JSON serializable

The dict can look like this:

annotation =
{'id': 1,
 'image_id': '0002',
 'category_id': 40,
 'segmentation': {'size': [480, 640],
  'counts': b'^cb45f>6K5J6J501O001O1O002N1O2N1O1O00O1000001O000000000000000000000000000O10000010O001O001O000000000000000000000001O003M5K4L4K5L5K4L4LQmj3'},
 'area': 1923.0,
 'bbox': [312, 347, 64, 37],
 'iscrowd': 0}

where the value in 'counts' is the bytes-type.

There is a way to solve the problem by converting the bytes to a string by applying:

annotation['segmentation']['counts'] = str(annotation['segmentation']['counts'])

But here is the problem: the JSON file would look like this (note that there are quotation marks (“) around the bytes):

{
    "id": 1,
    "image_id": "0002",
    "category_id": 40,
    "segmentation": {
        "size": [
            480,
            640
        ],
    "counts": "b'^cb45f>6K5J6J501O001O1O002N1O2N1O1O00O1000001O000000000000000000000000000O10000010O001O001O000000000000000000000001O003M5K4L4K5L5K4L4LQmj3'"
    },
    "area": 1923.0,
    "bbox": [
        312,
        347,
        64,
        37
    ],
    "iscrowd": 0
}

The COCO format needs a form like this though (no quotation marks around the bytes):

{
    "id": 1,
    "image_id": "0002",
    "category_id": 40,
    "segmentation": {
        "size": [
            480,
            640
        ],
    "counts": b'^cb45f>6K5J6J501O001O1O002N1O2N1O1O00O1000001O000000000000000000000000000O10000010O001O001O000000000000000000000001O003M5K4L4K5L5K4L4LQmj3'
    },
    "area": 1923.0,
    "bbox": [
        312,
        347,
        64,
        37
    ],
    "iscrowd": 0
}

Is there a way to write the JSON file in the correct format?