Package com.itextpdf.pdfocr.onnxtr.orientation

Class OnnxOrientationPredictor

java.lang.Object

com.itextpdf.pdfocr.onnxtr.AbstractOnnxPredictor<BufferedImage, TextOrientation>

com.itextpdf.pdfocr.onnxtr.orientation.OnnxOrientationPredictor

All Implemented Interfaces:

IPredictor<BufferedImage,TextOrientation>, IOrientationPredictor, AutoCloseable

       public class OnnxOrientationPredictor extends AbstractOnnxPredictor<BufferedImage,TextOrientation> implements IOrientationPredictor
      

A crop orientation predictor implementation, which is using ONNX Runtime and its ML models to figure out, how text is oriented in a cropped image of text.

Constructor Summary

Constructors

Constructor	Description
OnnxOrientationPredictor(OnnxOrientationPredictorProperties properties)	Creates a crop orientation predictor with the specified properties.

Method Summary

Modifier and Type	Method	Description
protected List<TextOrientation>	fromOutputBuffer(List<BufferedImage> inputBatch, FloatBufferMdArray outputBatch)	Converts ONNX runtime model batched output MD-array buffer to a list of predictor outputs.
OnnxOrientationPredictorProperties	getProperties()	Returns the crop orientation predictor properties.
static OnnxOrientationPredictor	mobileNetV3(String modelPath)	Creates a new crop orientation predictor using an existing pre-trained MobileNetV3 model, stored on disk.
protected FloatBufferMdArray	toInputBuffer(List<BufferedImage> batch)	Converts predictor inputs to an ONNX runtime model batched input MD-array buffer.

Methods inherited from class com.itextpdf.pdfocr.onnxtr.AbstractOnnxPredictor

close, predict

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface java.lang.AutoCloseable

close

Methods inherited from interface com.itextpdf.pdfocr.onnxtr.IPredictor

predict, predict

Constructor Details

OnnxOrientationPredictor

             public OnnxOrientationPredictor
             (OnnxOrientationPredictorProperties properties)
            

Creates a crop orientation predictor with the specified properties.

Parameters:

properties - properties of the predictor

Method Details

mobileNetV3

             public static OnnxOrientationPredictor mobileNetV3
             (String modelPath)
            

Creates a new crop orientation predictor using an existing pre-trained MobileNetV3 model, stored on disk. This is the only crop orientation model architecture available in OnnxTR.

This can be used to load the following models from OnnxTR:

• mobilenet_v3_small_crop_orientation
• mobilenet_v3_small_crop_orientation (8-bit quantized)

Parameters:

modelPath - path to the pre-trained model

Returns:

a new predictor with the MobileNetV3 model loaded

getProperties

             public OnnxOrientationPredictorProperties getProperties()
            

Returns the crop orientation predictor properties.

Returns:

the crop orientation predictor properties

toInputBuffer

             protected FloatBufferMdArray toInputBuffer
             (List<BufferedImage> batch)
            

Converts predictor inputs to an ONNX runtime model batched input MD-array buffer.

Specified by:

toInputBuffer in class AbstractOnnxPredictor<BufferedImage,TextOrientation>

Parameters:

batch - batch of raw predictor inputs

Returns:

batched model input MD-array buffer

fromOutputBuffer

             protected List<TextOrientation> fromOutputBuffer
             (List<BufferedImage> inputBatch, FloatBufferMdArray outputBatch)
            

Converts ONNX runtime model batched output MD-array buffer to a list of predictor outputs.

Specified by:

fromOutputBuffer in class AbstractOnnxPredictor<BufferedImage,TextOrientation>

Parameters:

inputBatch - list of raw predictor inputs, matching the output

outputBatch - batched model output MD-array buffer

Returns:

a list of predictor output