Package com.itextpdf.pdfocr.onnxtr

Class AbstractOnnxPredictorR>

java.lang.Object

com.itextpdf.pdfocr.onnxtr.AbstractOnnxPredictorR>

Type Parameters:

T - predictor input type

R - predictor output type

All Implemented Interfaces:

IPredictorR>, AutoCloseable

Direct Known Subclasses:

OnnxDetectionPredictor, OnnxOrientationPredictor, OnnxRecognitionPredictor

       public abstract class AbstractOnnxPredictorR> extends Object implements IPredictorR>
      

Abstract predictor, based on models running over ONNX runtime.

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	AbstractOnnxPredictor(String modelPath, OnnxInputProperties inputProperties, long[] outputShape)	Creates a new abstract predictor.

Method Summary

Modifier and Type	Method	Description
void	close()
protected abstract List<R>	fromOutputBuffer(List<T> inputBatch, FloatBufferMdArray outputBatch)	Converts ONNX runtime model batched output MD-array buffer to a list of predictor outputs.
Iterator<R>	predict(Iterator<T> inputs)	Performs prediction on a sequence of input items.
protected abstract FloatBufferMdArray	toInputBuffer(List<T> batch)	Converts predictor inputs to an ONNX runtime model batched input MD-array buffer.

Methods inherited from class java.lang.Object

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

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

predict

Constructor Details

AbstractOnnxPredictor

             protected AbstractOnnxPredictor
             (String modelPath, OnnxInputProperties inputProperties, long[] outputShape)
            

Creates a new abstract predictor.

If the specified model does not match input and output properties, it will throw an exception.

Parameters:

modelPath - path to the ONNX runtime model to load

inputProperties - expected input properties of a model

outputShape - expected shape of the output. -1 entries mean that the dimension can be of any size (ex. batch size)

Method Details

predict

             public Iterator<R> predict
             (Iterator<T> inputs)
            

Performs prediction on a sequence of input items.

This method consumes the provided Iterator of inputs and produces an Iterator of outputs, typically yielding one result per input item.

Specified by:

predict in interface IPredictor<T,R>

Parameters:

inputs - an Iterator over the input items to be processed

Returns:

an Iterator over the predicted output items

close

             public void close()
            

Specified by:

close in interface AutoCloseable

toInputBuffer

             protected abstract FloatBufferMdArray toInputBuffer
             (List<T> batch)
            

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

Parameters:

batch - batch of raw predictor inputs

Returns:

batched model input MD-array buffer

fromOutputBuffer

             protected abstract List<R> fromOutputBuffer
             (List<T> inputBatch, FloatBufferMdArray outputBatch)
            

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

Parameters:

inputBatch - list of raw predictor inputs, matching the output

outputBatch - batched model output MD-array buffer

Returns:

a list of predictor output