Response 400 from Google Vision API OCR with a bas

I've read How to use the Google Vision API for text detection from base64 encoded image? but it doesn't help at all. Cloud client library is undesirable for me because I am doing many image processing (e.g. rotating, cropping, resizing, etc.) before and during OCR. Saving them as new files and re-read them as inputs of Google Vision API is rather inefficient.

Hence, I went check the documentation of posting requests directly:

Using Python to send requests
Base64 Encoding
Optical character recognition (OCR),

and here are minimum codes to make the failure:

import base64
import requests
import io

# Read the image file and transform it into a base64 string
with io.open("photos/foo.jpg", 'rb') as image_file:
    image = image_file.read()
content = base64.b64encode(image)

# Prepare the data for request
# Format copied from https://cloud.google.com/vision/docs/ocr
sending_request = {
  "requests": [
    {
      "image": {
        "content": content
      },
      "features": [
        {
          "type": "TEXT_DETECTION"
        }
      ]
    }
  ]
}

# Send the request and get the response
# Format copied from https://cloud.google.com/vision/docs/using-python
response = requests.post(
    url='https://vision.googleapis.com/v1/images:annotate?key={}'.format(API_KEY),
    data=sending_request,
    headers={'Content-Type': 'application/json'}
)

# Then get 400 code
response
# <Response [400]>
print(response.text)
{
  "error": {
    "code": 400,
    "message": "Invalid JSON payload received. Unexpected token.\nrequests=image&reque\n^",
    "status": "INVALID_ARGUMENT"
  }
}

I went to my console and see there are indeed request errors for google.cloud.vision.v1.ImageAnnotator.BatchAnnotateImages, but I don't know what happened. Is it because the wrong format of sent data in requests.post?

The error, "message": "Invalid JSON payload received. Unexpected token.\nrequests=image&reque\n^", states you are passing a non-json format which is required to be a json. So, you should convert it to json and pass it to the request, as shown below.

response = requests.post(
url='https://vision.googleapis.com/v1/images:annotate?key={}'.format(API_KEY),
# import json module
# dumps the object to JSON
data=json.dumps(sending_request), 
headers={'Content-Type': 'application/json'}

It will trigger typeError: Object of type 'bytes' is not JSON serializable at the line of json.dumps([sending_request]) because you are not decoding the b64encode image. So, do this first and send the request

content = base64.b64encode(image).decode('UTF-8')