Package ai.djl.basicmodelzoo.cv.object_detection.yolo

Class YOLOV3

java.lang.Object
ai.djl.nn.AbstractBaseBlock
ai.djl.nn.AbstractBlock
ai.djl.basicmodelzoo.cv.object_detection.yolo.YOLOV3
All Implemented Interfaces:
ai.djl.nn.Block
public final class YOLOV3 extends ai.djl.nn.AbstractBlock
YOLOV3 contains a generic implementation of yolov3 (Original author bubbliiiing).

Yolov3 is a fast and accurate model for ObjectDetection tasks.

See Also:

  • Nested Class Summary

    Nested Classes
    Modifier and Type
    Class
    Description
    static final class 
    YOLOV3.Builder
    The Builder to construct a YOLOV3 object.

  • Field Summary

    Fields inherited from class ai.djl.nn.AbstractBlock

    children, parameters

    Fields inherited from class ai.djl.nn.AbstractBaseBlock

    inputNames, inputShapes, outputDataTypes, version

  • Method Summary

    Modifier and Type
    Method
    Description
    static ai.djl.nn.Block
    basicBlock(int filters, float batchNormMomentum, float leakyAlpha)
    Builds a Block that a basic residual block unit used in DarkNet53.
    static YOLOV3.Builder
    builder()
    Creates a builder to build a YOLOV3.
    static ai.djl.nn.Block
    convolutionBlock(int filters, int kernel, float batchNormMomentum, float leakyAlpha)
    Builds a Block that represents a conv-bn-leakyRelu unit for darkNet53.
    protected ai.djl.ndarray.NDList
    forwardInternal(ai.djl.training.ParameterStore parameterStore, ai.djl.ndarray.NDList inputs, boolean training, ai.djl.util.PairList<String,Object> params)
    ai.djl.ndarray.types.Shape[]
    getOutputShapes(ai.djl.ndarray.types.Shape[] inputShapes)
    void
    initializeChildBlocks(ai.djl.ndarray.NDManager manager, ai.djl.ndarray.types.DataType dataType, ai.djl.ndarray.types.Shape... inputShapes)
    static ai.djl.nn.Block
    makeLastLayers(int filtersIn, int filtersOut, float batchNormMomentum, float leakyAlpha)
    Builds a Block that represents the feature head in yolov3.
    static ai.djl.nn.Block
    makeLayer(int filters, int repeats, float batchNormMomentum, float leakyAlpha)
    Creates repeated Residual Blocks used in DarkNet53.
    static ai.djl.nn.Block
    makeOutputLayers(int filtersOut, int outClass, float batchNormMomentum, float leakyAlpha)
    Builds a Block that represents the output layer of yolov3.
    static ai.djl.nn.Block
    upSampleBlockNearest()
    Builds a Block that represents an upSampleLayer(the nearest mode) for yolov3.

    Methods inherited from class ai.djl.nn.AbstractBlock

    addChildBlock, addChildBlock, addChildBlockSingleton, addParameter, getChildren, getDirectParameters

    Methods inherited from class ai.djl.nn.AbstractBaseBlock

    beforeInitialize, cast, clear, describeInput, forward, forward, forwardInternal, getInputShapes, getOutputDataTypes, getParameters, initialize, isInitialized, loadMetadata, loadParameters, prepare, readInputShapes, saveInputShapes, saveMetadata, saveParameters, setInitializer, setInitializer, setInitializer, toString

    Methods inherited from class java.lang.Object

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

    Methods inherited from interface ai.djl.nn.Block

    forward, freezeParameters, freezeParameters, getOutputShapes

  • Method Details

    • upSampleBlockNearest

      public static ai.djl.nn.Block upSampleBlockNearest()
      Builds a Block that represents an upSampleLayer(the nearest mode) for yolov3.
      Returns:
      a Block that represent an upSampleLayer for yolov3

    • convolutionBlock

      public static ai.djl.nn.Block convolutionBlock(int filters, int kernel, float batchNormMomentum, float leakyAlpha)
      Builds a Block that represents a conv-bn-leakyRelu unit for darkNet53.
      Parameters:
      filters - the number of filters for conv
      kernel - the kernel size for conv
      batchNormMomentum - the momentum for batchNorm layer
      leakyAlpha - the alpha for leakyRelu activation
      Returns:
      a Block that represents a conv-bn-leakyRelu unit for darkNet53

    • makeLastLayers

      public static ai.djl.nn.Block makeLastLayers(int filtersIn, int filtersOut, float batchNormMomentum, float leakyAlpha)
      Builds a Block that represents the feature head in yolov3.
      Parameters:
      filtersIn - the number of input filters
      filtersOut - the number of output filters
      batchNormMomentum - the momentum of batchNorm layer
      leakyAlpha - the alpha value for leakyRelu activation
      Returns:
      a Block that represents the feature head in yolov3.

    • makeOutputLayers

      public static ai.djl.nn.Block makeOutputLayers(int filtersOut, int outClass, float batchNormMomentum, float leakyAlpha)
      Builds a Block that represents the output layer of yolov3.
      Parameters:
      filtersOut - the number of output filters
      outClass - the number of output classes
      batchNormMomentum - the momentum for batchNorm layer
      leakyAlpha - the alpha for leakyRelu activation
      Returns:
      a Block that represents the output layer of yolov3.

    • forwardInternal

      protected ai.djl.ndarray.NDList forwardInternal(ai.djl.training.ParameterStore parameterStore, ai.djl.ndarray.NDList inputs, boolean training, ai.djl.util.PairList<String,Object> params)
      Specified by:
      forwardInternal in class ai.djl.nn.AbstractBaseBlock

    • getOutputShapes

      public ai.djl.ndarray.types.Shape[] getOutputShapes(ai.djl.ndarray.types.Shape[] inputShapes)

    • initializeChildBlocks

      public void initializeChildBlocks(ai.djl.ndarray.NDManager manager, ai.djl.ndarray.types.DataType dataType, ai.djl.ndarray.types.Shape... inputShapes)
      Overrides:
      initializeChildBlocks in class ai.djl.nn.AbstractBaseBlock

    • basicBlock

      public static ai.djl.nn.Block basicBlock(int filters, float batchNormMomentum, float leakyAlpha)
      Builds a Block that a basic residual block unit used in DarkNet53.
      Parameters:
      filters - the output filter of the Convolutional Layer
      batchNormMomentum - the momentum used for computing batchNorm
      leakyAlpha - the alpha used in LeakyRelu Function
      Returns:
      a basic residual block unit used in DarkNet53

    • makeLayer

      public static ai.djl.nn.Block makeLayer(int filters, int repeats, float batchNormMomentum, float leakyAlpha)
      Creates repeated Residual Blocks used in DarkNet53.
      Parameters:
      filters - the output filters of the final Convolutional Layer
      repeats - the repeat times of a residual unit
      batchNormMomentum - the momentum used for computing batchNorm
      leakyAlpha - the alpha used in LeakyRelu Function
      Returns:
      several repeats of a residual block

    • builder

      public static YOLOV3.Builder builder()
      Creates a builder to build a YOLOV3.
      Returns:
      a new builder