package onnx

  1. Alphabetic
  1. Public
  2. All

Type Members

  1. class ImageFeaturizer extends Transformer with HasInputCol with HasOutputCol with Wrappable with ComplexParamsWritable with SynapseMLLogging

    The ImageFeaturizer relies on a ONNX model to do the featurization.

    The ImageFeaturizer relies on a ONNX model to do the featurization. One can set this model using the setOnnxModel parameter with a model you create yourself, or setModel to get a predefined named model from the ONNXHub.

    The ImageFeaturizer takes an input column of images (the type returned by the ImageReader), and automatically resizes them to fit the ONNXModel's inputs. It then feeds them through a pre-trained ONNX model.

  2. case class ONNXExtraPorts(features: Seq[ONNXShape]) extends Product with Serializable
  3. class ONNXHub extends Logging
  4. case class ONNXIOPorts(inputs: Seq[ONNXShape], outputs: Seq[ONNXShape]) extends Product with Serializable
  5. case class ONNXMetadata(modelSha: Option[String], modelBytes: Option[Long], tags: Option[Seq[String]], ioPorts: Option[ONNXIOPorts], extraPorts: Option[ONNXExtraPorts], modelWithDataPath: Option[String], modelWithDataSha: Option[String], modelWithDataBytes: Option[Long]) extends Product with Serializable
  6. class ONNXModel extends Transformer with ComplexParamsWritable with ONNXModelParams with Wrappable with SynapseMLLogging
  7. case class ONNXModelInfo(model: String, modelPath: String, onnxVersion: String, opsetVersion: Int, metadata: ONNXMetadata) extends Product with Serializable
  8. trait ONNXModelParams extends Params with HasMiniBatcher with HasFeedFetchDicts
  9. case class ONNXShape(name: String, shape: Seq[Either[Option[String], Int]], type: Option[String]) extends Product with Serializable

Value Members

  1. object ImageFeaturizer extends ComplexParamsReadable[ImageFeaturizer] with Serializable
  2. object ONNXHub
  3. object ONNXHubJsonProtocol extends DefaultJsonProtocol
  4. object ONNXModel extends ComplexParamsReadable[ONNXModel] with Serializable

    Object model for an ONNX model: OrtSession

    Object model for an ONNX model: OrtSession

    |-InputInfo: Map[String, NodeInfo]

    |-OutputInfo: Map[String, NodeInfo] OrtSession is the entry point for the object model. Most importantly it defines the InputInfo and OutputInfo maps. ------------------------------------ NodeInfo

    |-name: String

    |-info: ValueInfo Each NodeInfo is a name and ValueInfo tuple. ValueInfo has three implementations, explained below. ------------------------------------ TensorInfo extends ValueInfo

    |-shape: Array[Long]

    |-type: OnnxJavaType TensorInfo is the most common type of ValueInfo. It defines the type of the tensor elements, and the shape. The first dimension of the tensor is assumed to be the batch size. For example, FLOAT[-1, 3, 224, 224] could represent a unlimited batch size * 3 channels * 224 height * 224 width tensor, where each element is a float. ------------------------------------ SequenceInfo extends ValueInfo

    |-sequenceOfMaps: Boolean

    |-sequenceType: OnnxJavaType

    |-mapInfo: MapInfo

    |-length: Int SequenceInfo can be a sequence of values (value type specified by sequenceType) if sequenceOfMaps is false, or a sequence of MapInfo if sequenceOfMaps is true. Sequence of MapInfo is usually used for ZipMap type of output, where the sequence represent the batch, and each MapInfo represents probability or logits outcome per class for each observation. ------------------------------------ MapInfo extends ValueInfo

    |-keyType: OnnxJavaType

    |-valueType: OnnxJavaType

    |-size: Int MapInfo defines keyType, valueType and size. It is usually used inside SequenceInfo.

  5. object ONNXRuntime extends Logging

    ONNXRuntime: A wrapper around the ONNX Runtime (ORT)

  6. object ONNXUtils