Package org.anchoranalysis.plugin.onnx.bean.object.segment.decode.instance.maskrcnn

Class DecodeMaskRCNN

Object
AnchorBean<DecodeInstanceSegmentation<ai.onnxruntime.OnnxTensor>>
DecodeInstanceSegmentation<ai.onnxruntime.OnnxTensor>
DecodeMaskRCNN
public class DecodeMaskRCNN extends DecodeInstanceSegmentation<ai.onnxruntime.OnnxTensor>
Decodes the inference output from a Mask-RCNN implementation.

It is designed to work with accompanying MaskRCNN-10.onnx in resources, which expects an image of size 1088x800 (width x height) and may throw an error if the input-size is different than this.

The ONNX file was obtained from this GitHub source, which also describes its inputs and outputs.

This issue may also be relevant: it mentions an error message that occurs when a different sized input to the above is used.

Author:
Owen Feehan

  • Constructor Details

    • DecodeMaskRCNN

      public DecodeMaskRCNN()

  • Method Details

    • expectedOutputs

      public List<String> expectedOutputs()
      Description copied from class: DecodeInstanceSegmentation
      Ordered names of the tensors we are interested in processing, as outputted from inference.
      Specified by:
      expectedOutputs in class DecodeInstanceSegmentation<ai.onnxruntime.OnnxTensor>
      Returns:
      the list of names, as above.

    • decode

      public List<LabelledWithConfidence<MultiScaleObject>> decode(List<ai.onnxruntime.OnnxTensor> inferenceOutput, ImageInferenceContext context) throws OperationFailedException
      Description copied from class: DecodeInstanceSegmentation
      Decodes the output tensors from inference into ObjectMasks with confidence and labels.

      The created ObjectMasks should match unscaledDimensions in size.

      Specified by:
      decode in class DecodeInstanceSegmentation<ai.onnxruntime.OnnxTensor>
      Parameters:
      inferenceOutput - the tensors that are the result of the inference.
      context - the context in which the inference is occurring.
      Returns:
      a newly created list of objects, with associated confidence, and labels, that matches unscaledDimensions in size.
      Throws:
      OperationFailedException - if it cannot be decoded successfully.

    • getMinConfidence

      public float getMinConfidence()
      Only proposals outputted from the model with a score greater or equal to this threshold are considered.

    • setMinConfidence

      public void setMinConfidence(float minConfidence)
      Only proposals outputted from the model with a score greater or equal to this threshold are considered.

    • getMinMaskValue

      public float getMinMaskValue()
      Only voxels with a value greater or equal to this threshold are considered as part of the mask.

    • setMinMaskValue

      public void setMinMaskValue(float minMaskValue)
      Only voxels with a value greater or equal to this threshold are considered as part of the mask.

    • getInterpolator

      public Interpolator getInterpolator()
      The interpolator to use for scaling images.

    • setInterpolator

      public void setInterpolator(Interpolator interpolator)
      The interpolator to use for scaling images.