themantalope / executor-image-paddlepaddle-ocr

PaddlepaddleOCR

PaddlepaddleOCR wraps the models from paddlepaddleOCR

PaddlepaddleOCR executor receives Documents with uri attribute. Each Document's uri represents the path of an image. This executor will read the image and extract bounding boxes surrouding text, the text itself and the confidence score.

Every result will be saved inside a document where text attribute will be the extracted text, weight attribute will be the confidence score and we add a coordinates tag containing the bounding box coordinates. The result is finally saved as a chunk in the original document containing the image.

Usage

Use the prebuilt images from Jina Hub in your Python code, add it to your Flow and extract text from an image:

from jina import Flow
	
f = Flow().add(uses='jinahub+docker://PaddlepaddleOCR')

docs = DocumentArray()
doc = Document(uri='/your/image/path')
docs.append(doc)

def print_results(resp):
    """
    Function to print the extracted text from the response after
    applying OCR to the input images

    :resp: response resulting for the executor
    """

    for doc in resp.docs:
        for chunk in doc.chunks:
            print(chunk.text)
            
with f:
    f.post(on='/extract', inputs=docs, on_done=print_results)

Returns

Document with text field filled with an str, weight field filled with float32 and coordinates filled with list of lists containing float32 representing the bounding box.

GPU usage

This executor also offers a GPU version. To use it, make sure to pass 'use_gpu'=True, as the initialization parameter, and gpus='all' when adding the containerized Executor to the Flow. See the Executor on GPU section of Jina documentation for more details.

Here's how you would modify the example above to use a GPU

from jina import Flow
	
f = Flow().add(
    uses='jinahub+docker://PaddlepaddleOCR',
    uses_with={'use_gpu':True},
    gpus='all
    )

Reference

PaddlepaddleOCR repository