Setting the Parameters

Specifying the Execution Settings

Depending on the computer used, you can greatly improve the performance by properly setting these deep learning execution settings.

EasyDeepOCR is based on eVision’s deep learning tools.

  • Use the method ETextReader::SetExecutionSettings .NET Python to set the execution settings of these tools.

  • It is not mandatory to set these execution settings. If they are not set, EasyDeepOCR tries to use the OpenVINO engine if available, otherwise it uses the default engine.

Warning

Changing the execution settings resets the initialization. You must call ETextReader::Initialize .NET Python afterward.

Warning

When using the TensorRT engine, the initialization may take a few tens of minutes, depending on the hardware configuration.

Building a Sample List

The class ESampleList .NET Python contains all the information about a “customer use case”.

The best way to create an ESampleList .NET Python is through a graphical interface because you need to annotate the “texts of interest”. eVision Studio provides a widget to perform this annotation easily. Alternatively, you can also create the ESampleList .NET Python using the API.

Using eVision Studio

../../../_images/building_a_sample_liste.webp

The eVision Studio widget for the edition of an ESampleList .NET Python

To manage your ESampleList .NET Python:

  1. Open a file explorer to add one or several images to the batch.

    • An ESample is created for each image added to the batch.

  2. Display the list of the ESample .NET Python. The ESample .NET Python currently selected is highlighted.

  3. Create a new EText .NET Python and add it to the selected ESample .NET Python.

  4. Display the list of EText .NET Python for the selected ESample .NET Python. The EText .NET Python currently selected is highlighted.

  5. Create a new ETextLine .NET Python and add it to the selected EText .NET Python.

  6. Display the list of ETextLine .NET Python for the selected EText .NET Python.

  7. Display the image of the selected ESample .NET Python, overlaid with the ETextLine .NET Python of the selected EText .NET Python.

Using the API

The creation of the ESampleList .NET Python is illustrated in the code snippet below.

#include <eVision/EasyOcr/EasyDeepOCR/ESampleList.h>

using namespace eVision;
namespace EDO = eVision::EasyDeepOCR;
// Create an ESampleList
EDO::ESampleList sampleList;
// Create an ESample vector
std::vector<EDO::ESample> samples(1);
// Fill the ESample vector
for (EDO::ESample& sample : samples)
{
  // Set the image
  sample.SetImagePath("image.tif");
  // Create an EText vector 
  std::vector<EDO::EText> texts(1);
  // Fill the EText vector
  for (EDO::EText& text : texts)
  {
    // Create an ETextLine vector
    std::vector<EDO::ETextLine> textLines(1);
    // Fill the ETextLine vector
    for (EDO::ETextLine& textLine : textLines)
    {
      // Adjust the ERectangle around the text line
      textLine.SetRectangle(ERectangle());
      // Annotate the text line
      textLine.SetString("");
    }
    // Set the ETextLine vector
    text.SetLines(textLines);
  }
  // Set the EText vector
  sample.SetTexts(texts);
}
// Set the ESample vector
sampleList.SetSamples(samples);
// ...

using eVision;
using EDO = eVision.EasyDeepOCR;

// Create an ESampleList
EDO.ESampleList sampleList = new EDO.ESampleList();
// Create an ESample vector
EDO.ESample[] samples = new EDO.ESample[1];
// Fill the ESample vector
foreach (EDO.ESample sample in samples) 
{
  // Set the image
  sample.ImagePath = "image.tif";
  // Create an EText vector 
  EDO.EText[] texts = new EDO.EText[1];
  // Fill the EText vector
  foreach (EDO.EText text in texts)
  {
    // Create an ETextLine vector
    EDO.ETextLine[] textLines = new EDO.ETextLine[1];
    // Fill the ETextLine vector
    foreach (EDO.ETextLine textLine in textLines)
    {
      // Adjust the ERectangle around the text line
      textLine.Rectangle = new ERectangle();
      // Annotate the text line
      textLine.String = "";
    }
    // Set the ETextLine vector
    text.Lines = textLines;
  }
  // Set the EText vector
  sample.Texts = texts;
}
// Set the ESample vector
sampleList.Samples = samples;
// ...

import eVision
from eVision import EasyDeepOCR as EDO
# Create an ESampleList
sampleList = EDO.ESampleList()
# Create an ESample vector
samples = [EDO.ESample()]
# Fill the ESample vector
for sample in samples: 
  # Set the image
  sample.ImagePath = "image.tif"
  # Create an EText vector 
  texts = [EDO.EText()]
  # Fill the EText vector
  for text in texts:
    # Create an ETextLine vector
    textLines = [EDO.ETextLine()]
    # Fill the ETextLine vector
    for textLine in textLines:
      # Adjust the ERectangle around the text line
      textLine.Rectangle = eVision.ERectangle()
      # Annotate the text line
      textLine.String = ""
    # Set the ETextLine vector
    text.Lines = textLines
  # Set the EText vector
  sample.Texts = texts
# Set the ESample vector
sampleList.Samples = samples

You can easily modify this code snippet to include more ESample .NET Python, EText .NET Python and ETextLine .NET Python.

Warning

EasyDeepOCR expects that the rectangle of the ETextLine .NET Python is tightly fitted around the text. There must be no margin between the text line and the border of the rectangle.