Package apple.metalperformanceshaders

Class MPSCNNYOLOLossDescriptor

  • All Implemented Interfaces:
    NSCopying, NSObject
    public class MPSCNNYOLOLossDescriptor
extends NSObject
implements NSCopying
    MPSCNNYOLOLossDescriptor [@dependency] This depends on Metal.framework. The MPSCNNYOLOLossDescriptor specifies a loss filter descriptor that is used to create a MPSCNNLoss filter. The MPSCNNYOLOLoss is a filter that has been specialized for object detection tasks and follows a specific layout for the feature-channels of the input, output, weight and label data. The layout of the data within the feature-channels is as follows: Each anchorbox uses ( 2+2+1 + numberOfClasses = 5 + numberOfClasses ) feature channels. Therefore the total number of feature channels used is: (5 + numberOfClasses) * numberOfAnchorBoxes. The first feature channel for anchorbox index 'anchorIdx' is at fcIndex = (5 + numberOfClasses) * anchorIdx, and the feature channels within each anchorbox are stored in the layout: 'XYWHCFFFFFF...', where (XY) are the so-called raw x and y coordinates of the bounding box within each gridcell and (WH) are the corresponding width and height. 'C' signifies a confidence for having an object in the cell and FFFFF... are the feature channel values for each class of object to be classified in the object detector. The YOLO-loss filter works by operating mostly independently on each anchorbox: * The XY-channels of the inputs are first transformed to relative XY-values by applying the sigmoid-neuron on them, after which they are passed through the loss function defined by @ref XYLossDescriptor, which is typically chosen to be the @ref MPSCNNLossTypeMeanSquaredError type loss function. * The WH-channels contain the raw width and height of the bounding box and they are operated with the loss function defined by @ref WHLossDescriptor, which is typically of type @ref MPSCNNLossTypeHuber. * The C-channel contains the confidence value of having an object in the bounding box and it is operated by the loss function defined by @ref confidenceLossDescriptor, which is typically chosen to be [@ref] MPSCNNLossTypeSigmoidCrossEntropy. * The FFFFF... (number of channels is number of classes) channels contains the raw feature channels for object classes, used to identify which objects are the most probable ones in the bounding box and these channels are passed through the loss function defined by @ref classesLossDescriptor, which in typical cases is of the type @ref MPSCNNLossTypeSoftMaxCrossEntropy. For details on how to set up the label values and anchorboxes see https://arxiv.org/abs/1612.08242

    • Constructor Detail

      • MPSCNNYOLOLossDescriptor

        protected MPSCNNYOLOLossDescriptor​(org.moe.natj.general.Pointer peer)

    • Method Detail

      • WHLossDescriptor

        public MPSCNNLossDescriptor WHLossDescriptor()
        [@property] WHLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • XYLossDescriptor

        public MPSCNNLossDescriptor XYLossDescriptor()
        [@property] XYLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • accessInstanceVariablesDirectly

        public static boolean accessInstanceVariablesDirectly()

      • allocWithZone

        public static java.lang.Object allocWithZone​(org.moe.natj.general.ptr.VoidPtr zone)

      • anchorBoxes

        public NSData anchorBoxes()
        [@property] anchorBoxes NSData containing the width and height for numberOfAnchorBoxes anchor boxes This NSData should have 2 float values per anchor box which represent the width and height of the anchor box. [@code] typedef struct anchorBox{ float width; float height; }anchorBox; anchorBox_t gAnchorBoxes[MAX_NUM_ANCHOR_BOXES] = { {.width = 1.f, .height = 2.f}, {.width = 1.f, .height = 1.f}, {.width = 2.f, .height = 1.f}, }; NSData* labelsInputData = [NSData dataWithBytes: gAnchorBoxes length: MAX_NUM_ANCHOR_BOXES * sizeof(anchorBox)]; [@endcode]

      • automaticallyNotifiesObserversForKey

        public static boolean automaticallyNotifiesObserversForKey​(java.lang.String key)

      • cancelPreviousPerformRequestsWithTarget

        public static void cancelPreviousPerformRequestsWithTarget​(java.lang.Object aTarget)

      • cancelPreviousPerformRequestsWithTargetSelectorObject

        public static void cancelPreviousPerformRequestsWithTargetSelectorObject​(java.lang.Object aTarget,
                                                                         org.moe.natj.objc.SEL aSelector,
                                                                         java.lang.Object anArgument)

      • classFallbacksForKeyedArchiver

        public static NSArray<java.lang.String> classFallbacksForKeyedArchiver()

      • classForKeyedUnarchiver

        public static org.moe.natj.objc.Class classForKeyedUnarchiver()

      • classesLossDescriptor

        public MPSCNNLossDescriptor classesLossDescriptor()
        [@property] classesLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • cnnLossDescriptorWithXYLossTypeWHLossTypeConfidenceLossTypeClassesLossTypeReductionTypeAnchorBoxesNumberOfAnchorBoxes

        public static MPSCNNYOLOLossDescriptor cnnLossDescriptorWithXYLossTypeWHLossTypeConfidenceLossTypeClassesLossTypeReductionTypeAnchorBoxesNumberOfAnchorBoxes​(int XYLossType,
                                                                                                                                                             int WHLossType,
                                                                                                                                                             int confidenceLossType,
                                                                                                                                                             int classesLossType,
                                                                                                                                                             int reductionType,
                                                                                                                                                             NSData anchorBoxes,
                                                                                                                                                             long numberOfAnchorBoxes)
        Make a descriptor for a MPSCNNYOLOLoss object.
        Parameters:
        XYLossType - The type of spatial position loss filter.
        WHLossType - The type of spatial size loss filter.
        confidenceLossType - The type of confidence filter.
        classesLossType - The type of classes filter.
        reductionType - The type of a reduction operation to apply.
        anchorBoxes - This is an NSData which has an array of anchorBoxes defined as a struct{ float width; float height; };
        Returns:
        A valid MPSCNNYOLOLossDescriptor object or nil, if failure.

      • confidenceLossDescriptor

        public MPSCNNLossDescriptor confidenceLossDescriptor()
        [@property] confidenceLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • copyWithZone

        public java.lang.Object copyWithZone​(org.moe.natj.general.ptr.VoidPtr zone)
        Specified by:
        copyWithZone in interface NSCopying

      • debugDescription_static

        public static java.lang.String debugDescription_static()

      • description_static

        public static java.lang.String description_static()

      • hash_static

        public static long hash_static()

      • instanceMethodSignatureForSelector

        public static NSMethodSignature instanceMethodSignatureForSelector​(org.moe.natj.objc.SEL aSelector)

      • instancesRespondToSelector

        public static boolean instancesRespondToSelector​(org.moe.natj.objc.SEL aSelector)

      • isSubclassOfClass

        public static boolean isSubclassOfClass​(org.moe.natj.objc.Class aClass)

      • keyPathsForValuesAffectingValueForKey

        public static NSSet<java.lang.String> keyPathsForValuesAffectingValueForKey​(java.lang.String key)

      • maxIOUForObjectAbsence

        public float maxIOUForObjectAbsence()
        [@property] neg_iou If the prediction IOU with groundTruth is lower than this value we consider it a confident object absence, default is 0.3

      • minIOUForObjectPresence

        public float minIOUForObjectPresence()
        [@property] pos_iou If the prediction IOU with groundTruth is higher than this value we consider it a confident object presence, default is 0.7

      • new_objc

        public static java.lang.Object new_objc()

      • numberOfAnchorBoxes

        public long numberOfAnchorBoxes()
        [@property] numberOfAnchorBoxes number of anchor boxes used to detect object per grid cell

      • reductionType

        public int reductionType()
        [@property] reductionType ReductionType shared accross all losses (so they may generate same sized output)

      • rescore

        public boolean rescore()
        [@property] rescore Rescore pertains to multiplying the confidence groundTruth with IOU (intersection over union) of predicted bounding box and the groundTruth boundingBox. Default is YES

      • resolveClassMethod

        public static boolean resolveClassMethod​(org.moe.natj.objc.SEL sel)

      • resolveInstanceMethod

        public static boolean resolveInstanceMethod​(org.moe.natj.objc.SEL sel)

      • scaleClass

        public float scaleClass()
        [@property] scaleClass scale factor for no object classes loss and loss gradient default is 2.0

      • scaleNoObject

        public float scaleNoObject()
        [@property] scaleNoObject scale factor for no object confidence loss and loss gradient default is 5.0

      • scaleObject

        public float scaleObject()
        [@property] scaleObject scale factor for no object confidence loss and loss gradient default is 100.0

      • scaleWH

        public float scaleWH()
        [@property] scaleWH scale factor for WH loss and loss gradient default is 10.0

      • scaleXY

        public float scaleXY()
        [@property] scaleXY scale factor for XY loss and loss gradient default is 10.0

      • setAnchorBoxes

        public void setAnchorBoxes​(NSData value)
        [@property] anchorBoxes NSData containing the width and height for numberOfAnchorBoxes anchor boxes This NSData should have 2 float values per anchor box which represent the width and height of the anchor box. [@code] typedef struct anchorBox{ float width; float height; }anchorBox; anchorBox_t gAnchorBoxes[MAX_NUM_ANCHOR_BOXES] = { {.width = 1.f, .height = 2.f}, {.width = 1.f, .height = 1.f}, {.width = 2.f, .height = 1.f}, }; NSData* labelsInputData = [NSData dataWithBytes: gAnchorBoxes length: MAX_NUM_ANCHOR_BOXES * sizeof(anchorBox)]; [@endcode]

      • setClassesLossDescriptor

        public void setClassesLossDescriptor​(MPSCNNLossDescriptor value)
        [@property] classesLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • setConfidenceLossDescriptor

        public void setConfidenceLossDescriptor​(MPSCNNLossDescriptor value)
        [@property] confidenceLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • setMaxIOUForObjectAbsence

        public void setMaxIOUForObjectAbsence​(float value)
        [@property] neg_iou If the prediction IOU with groundTruth is lower than this value we consider it a confident object absence, default is 0.3

      • setMinIOUForObjectPresence

        public void setMinIOUForObjectPresence​(float value)
        [@property] pos_iou If the prediction IOU with groundTruth is higher than this value we consider it a confident object presence, default is 0.7

      • setNumberOfAnchorBoxes

        public void setNumberOfAnchorBoxes​(long value)
        [@property] numberOfAnchorBoxes number of anchor boxes used to detect object per grid cell

      • setReductionType

        public void setReductionType​(int value)
        [@property] reductionType ReductionType shared accross all losses (so they may generate same sized output)

      • setRescore

        public void setRescore​(boolean value)
        [@property] rescore Rescore pertains to multiplying the confidence groundTruth with IOU (intersection over union) of predicted bounding box and the groundTruth boundingBox. Default is YES

      • setScaleClass

        public void setScaleClass​(float value)
        [@property] scaleClass scale factor for no object classes loss and loss gradient default is 2.0

      • setScaleNoObject

        public void setScaleNoObject​(float value)
        [@property] scaleNoObject scale factor for no object confidence loss and loss gradient default is 5.0

      • setScaleObject

        public void setScaleObject​(float value)
        [@property] scaleObject scale factor for no object confidence loss and loss gradient default is 100.0

      • setScaleWH

        public void setScaleWH​(float value)
        [@property] scaleWH scale factor for WH loss and loss gradient default is 10.0

      • setScaleXY

        public void setScaleXY​(float value)
        [@property] scaleXY scale factor for XY loss and loss gradient default is 10.0

      • setVersion_static

        public static void setVersion_static​(long aVersion)

      • setWHLossDescriptor

        public void setWHLossDescriptor​(MPSCNNLossDescriptor value)
        [@property] WHLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • setXYLossDescriptor

        public void setXYLossDescriptor​(MPSCNNLossDescriptor value)
        [@property] XYLossDescriptor The type of a loss filter. This parameter specifies the type of a loss filter.

      • superclass_static

        public static org.moe.natj.objc.Class superclass_static()

      • version_static

        public static long version_static()

      • reduceAcrossBatch

        public boolean reduceAcrossBatch()
        [@property] reduceAcrossBatch If set to YES then the reduction operation is applied also across the batch-index dimension, ie. the loss value is summed over images in the batch and the result of the reduction is written on the first loss image in the batch while the other loss images will be set to zero. If set to NO, then no reductions are performed across the batch dimension and each image in the batch will contain the loss value associated with that one particular image. NOTE: If reductionType == MPSCNNReductionTypeNone, then this flag has no effect on results, that is no reductions are done in this case. NOTE: If reduceAcrossBatch is set to YES and reductionType == MPSCNNReductionTypeMean then the final forward loss value is computed by first summing over the components and then by dividing the result with: number of feature channels * width * height * number of images in the batch. The default value is NO.

      • setReduceAcrossBatch

        public void setReduceAcrossBatch​(boolean value)
        [@property] reduceAcrossBatch If set to YES then the reduction operation is applied also across the batch-index dimension, ie. the loss value is summed over images in the batch and the result of the reduction is written on the first loss image in the batch while the other loss images will be set to zero. If set to NO, then no reductions are performed across the batch dimension and each image in the batch will contain the loss value associated with that one particular image. NOTE: If reductionType == MPSCNNReductionTypeNone, then this flag has no effect on results, that is no reductions are done in this case. NOTE: If reduceAcrossBatch is set to YES and reductionType == MPSCNNReductionTypeMean then the final forward loss value is computed by first summing over the components and then by dividing the result with: number of feature channels * width * height * number of images in the batch. The default value is NO.