The Python OCR SDK supports the Cropper API.

Using the sample below, we are going to illustrate how to extract the data that we want using the OCR SDK.

Quick-Start

from mindee import Client, PredictResponse, product

# Init a new client
mindee_client = Client(api_key="my-api-key")

# Load a file from disk
input_doc = mindee_client.source_from_path("/path/to/the/file.ext")

# Load a file from disk and parse it.
# The endpoint name must be specified since it cannot be determined from the class.
result: PredictResponse = mindee_client.parse(product.CropperV1, input_doc)

# Print a summary of the API result
print(result.document)

# Print the document-level summary
# print(result.document.inference.prediction)

Output (RST):

########
Document
########
:Mindee ID: 149ce775-8302-4798-8649-7eda9fb84a1a
:Filename: default_sample.jpg

Inference
#########
:Product: mindee/cropper v1.0
:Rotation applied: No

Prediction
==========

Page Predictions
================

Page 0
------
:Document Cropper: Polygon with 26 points.
                   Polygon with 25 points.

Field Types

Standard Fields

These fields are generic and used in several products.

BaseField

Each prediction object contains a set of fields that inherit from the generic BaseField class.
A typical BaseField object will have the following attributes:

• value (Union[float, str]): corresponds to the field value. Can be None if no value was extracted.
• confidence (float): the confidence score of the field prediction.
• bounding_box ([Point, Point, Point, Point]): contains exactly 4 relative vertices (points) coordinates of a right rectangle containing the field in the document.
• polygon (List[Point]): contains the relative vertices coordinates (Point) of a polygon containing the field in the image.
• page_id (int): the ID of the page, always None when at document-level.
• reconstructed (bool): indicates whether an object was reconstructed (not extracted as the API gave it).

Note: A Point simply refers to a List of two numbers ([float, float]).

Aside from the previous attributes, all basic fields have access to a custom __str__ method that can be used to print their value as a string.

PositionField

The position field PositionField does not implement all the basic BaseField attributes, only bounding_box, polygon and page_id. On top of these, it has access to:

• rectangle ([Point, Point, Point, Point]): a Polygon with four points that may be oriented (even beyond canvas).
• quadrangle ([Point, Point, Point, Point]): a free polygon made up of four points.

Page-Level Fields

Some fields are constrained to the page level, and so will not be retrievable at document level.

Attributes

The following fields are extracted for Cropper V1:

Document Cropper

📄cropping (List[PositionField]): List of documents found in the image.

for page in result.document.inference.pages:
    for cropping_elem in page.prediction.cropping):
        print(cropping_elem.polygon)

Questions?

Join our Slack