Confidence score is not properly set when using Python [161429872]

WAI

Bug

Status Update

No update yet.

Description

rm...@google.com

created issue #1

Jul 16, 2020 12:10PM

Problem you have encountered: When using OCR to detect text in images [1], with Python client library, the confidence field is not set properly.
What you expected to happen: Confidence field should have the same value in the "Try the API" option and when performing the recognition with the python client library.
Steps to reproduce:

You can check this easily with any image. Run the following function with any image url:

from google.cloud import vision

def detect_text_uri(uri):
    client = vision.ImageAnnotatorClient()
    image = vision.types.Image()
    image.source.image_uri = uri

    response = client.text_detection(image=image)
    texts = response.text_annotations
    print('Texts:')

    for text in texts:
        print('\n"{}"'.format(text.description))

        vertices = (['({},{})'.format(vertex.x, vertex.y)
                    for vertex in text.bounding_poly.vertices])

        print('bounds: {}'.format(','.join(vertices)))
        print("confidence: {}".format(text.confidence))

    if response.error.message:
        raise Exception(
            '{}\nFor more info on error messages, check: '
            'https://cloud.google.com/apis/design/errors'.format(
                response.error.message))

You will see that confidences are 0.0. Then, use "Try the API" in [2] for the same image. You will see non zero confidences.

Other information (workarounds you have tried, documentation consulted, etc):

I have also checked that the same happens with a local image.

[1] https://cloud.google.com/vision/docs/ocr
[2] https://cloud.google.com/vision

Comments

ab...@google.com <ab...@google.com> #2Jul 18, 2020 02:52AM

Hi,
I did try to reproduce the issue but I did not get the same results.
I used the github detect.py example [1] which is the same as the one in Google documentation that you provided above.

1- I did try to test it locally where the photo is stored in my terminal local file using the following command line :
$ python test_document_detect.py document ./resources/wakeupcat.jpg

Which resulted to :
~~~~
Block confidence: 0.990000009537

Paragraph confidence: 0.990000009537
Word text: Wake (confidence: 0.990000009537)
Symbol: W (confidence: 0.990000009537)
Symbol: a (confidence: 1.0)
Symbol: k (confidence: 0.990000009537)
Symbol: e (confidence: 0.990000009537)
Word text: up (confidence: 1.0)
Symbol: u (confidence: 1.0)
Symbol: p (confidence: 1.0)
Word text: human (confidence: 0.990000009537)
Symbol: h (confidence: 0.990000009537)
Symbol: u (confidence: 0.990000009537)
Symbol: m (confidence: 0.990000009537)
Symbol: a (confidence: 1.0)
Symbol: n (confidence: 1.0)
Word text: ! (confidence: 0.959999978542)
Symbol: ! (confidence: 0.959999978542)
~~~~

2- Then I did try to test with providing a photo URI using the following command line :
$ python detect.py document-uri

https://www.researchgate.net/profile/Yaron_Kanza/publication/221326044/figure/fig1/AS:670048624574476@1536763449601/Labeled-tree-corresponding-to-the-document-fragment.png

Which resulted to :
~~~~
Block confidence: 0.990000009537

Paragraph confidence: 0.990000009537
Word text: inproceedings (confidence: 0.990000009537)
Symbol: i (confidence: 0.990000009537)
Symbol: n (confidence: 1.0)
Symbol: p (confidence: 0.990000009537)
Symbol: r (confidence: 0.990000009537)
Symbol: o (confidence: 1.0)
Symbol: c (confidence: 0.990000009537)
Symbol: e (confidence: 0.930000007153)
Symbol: e (confidence: 0.990000009537)
Symbol: d (confidence: 1.0)
Symbol: i (confidence: 1.0)
Symbol: n (confidence: 1.0)
Symbol: g (confidence: 1.0)
Symbol: s (confidence: 1.0)

Block confidence: 0.990000009537

Paragraph confidence: 0.990000009537
Word text: 2 (confidence: 0.990000009537)
Symbol: 2 (confidence: 0.990000009537)

Block confidence: 0.990000009537

Paragraph confidence: 0.990000009537
Word text: booktitle (confidence: 0.990000009537)
.
.
.
~~~~

For both cases we can have the level of confidence calculated with no issues, thank you.

[1]

https://github.com/GoogleCloudPlatform/python-docs-samples/blob/master/vision/cloud-client/detect/detect.py

ab...@google.com <ab...@google.com> #3Jul 25, 2020 03:46AM

Status: Won't Fix (Intended Behavior)

Hi,

This issue will be closed due to inactivity for last 7 days. If you face the same issue again please create a new case.

Thanks.

li...@gmail.com <li...@gmail.com> #4Sep 10, 2020 07:06PM

I actually ran into the same issue. I think the question asked about the API 'text_detection', while the answer was based on 'document_text_detection'.

Can you verify the issue with the 'text_detection' one?

Message last modified on Sep 10, 2020 07:15PM

zi...@gmail.com <zi...@gmail.com> #5Sep 11, 2020 12:24PM

Hi, I'm having the same issue with text_detection method. The python client gives confidence of 0 everywhere.

google-cloud-vision==1.0.0 python 3

pages = response.full_text_annotation.pages
for page in pages:
    for block in page.blocks:
        print('block confidence', block.confidence)
        
        for paragraph in block.paragraphs:
            print('paragraph confidence', paragraph.confidence)
            
            for word in paragraph.words:
                word_text = ''.join([symbol.text for symbol in word.symbols])
                
                print('Word text : {0} (confidence: {1})'.format(word_text, word.confidence))
                
                for symbol in word.symbols:
                    print('\tSymbol: {0} (confidence: {1})'.format(symbol.text, symbol.confidence))

Message last modified on Sep 16, 2020 03:22PM

hu...@gmail.com <hu...@gmail.com> #6Dec 8, 2020 03:14AM

Hi, I'm having the same issue with text_detection on google-cloud-vision (2.0.0). Please check the issues

sh...@gmail.com <sh...@gmail.com> #7Jan 7, 2021 11:20PM

Same issue for me as well. I am using the nodejs library: https://www.npmjs.com/package/@google-cloud/vision

const vision = require('@google-cloud/vision');
const client = new vision.ImageAnnotatorClient();
const [ocrResult] = await client.textDetection("/path/to/file.jpg");

The only confidence values returned correctly are the language detection ones.

Message last modified on Jan 7, 2021 11:22PM

fe...@old.stilingue.com.br <fe...@old.stilingue.com.br> #8Jan 12, 2021 02:36PM

Same issue for me, using python.

from google.cloud import vision
client = vision.ImageAnnotatorClient()
with io.open(path, 'rb') as image_file:
    content = image_file.read()
    
image = vision.Image(content=content)
response = client.text_detection(image=image)
texts = response.text_annotations
for text in texts:
    print(text.confidence) # always 0.0

ma...@outlook.be <ma...@outlook.be> #9Jan 15, 2021 07:35AM

Can confirm that issue is still active
Python 3.8.5, google-cloud-vision==2.0.0

Since the scanned files are not clean documents, I suppose implementing the code with the 'document_text_detection' function is not the best idea? Will test anyway

mi...@super.ai <mi...@super.ai> #10Sep 23, 2021 05:36PM

I still have the same issue as well. document_text_detection gives confidences but text_detection doesn't.
google-cloud-vision 2.4.2
python 3.7.11

Message last modified on Sep 23, 2021 05:36PM

sa...@gmail.com <sa...@gmail.com> #11Sep 23, 2021 05:41PM

I also have the same problem. Because this issue was closed and there were instructions to open a new one if necessary, I did that:

https://issuetracker.google.com/issues/200069044 - it has not been replied to yet, but I am leaving this comment here for others stumbling at this page with the same bug.

ve...@gmail.com <ve...@gmail.com> #12May 2, 2023 07:31AM

https://github.com/googleapis/python-vision/issues/266#issuecomment-973301113