public class MPSCNNPoolingGradient extends MPSCNNGradientKernel
Specifies the base class for computing the gradient of the pooling filters. The operation backpropagates a gradient vector using the chain rule.
Given the input gradient vector dL(x) = dL/d out(x), which is the derivative of the loss-function wrt. (original) pooling filter output the output gradient at position y (dL/d in(y)) is computed as follows:
dL/d in(y) = sum_x (dL/d out(x)) * (d out(x)/d in(y)), where
the sum runs over the input gradient pixels starting from primaryOffset extending to primaryOffset + sourceSize. Note here that we need a separate variable 'sourceSize' to specify which input gradients are included in the output gradient computation as this information cannot be deduced directly from the cliprect size due to fractional striding or simply because the user wants to examine a subset of the contributions to the gradients. In normal operation the sourceSize is specified as the cliprect.size of the forward pooling filter in order to compute the gradients for all outputs the forward direction produced and the primaryOffset is set to cliprect.origin of the original forward pooling operation for the same reason.
The cliprect property of the filter allows the user to send the gradients to a new location, which may not match the original forward pooling filter window locations: The index 'y' in the formula above refers to the pixel location in the secondary source, which is the source image of the original forward pooling filter and secondaryOffset specifies the center of the first pooling window as specified in MPSCNNPooling filter specification. The first (top leftmost) pixel written into the cliprect computes the derivative of the first pixel within the first pooling window that is contained within the secondary source image and subsequent values are defined by normal striding rules from secondary source to primary source. This means that typically the cliprect is set to fill the effective source area of the original forward operation, clamped to edges of the original source image, which in the normal case is the same size as the size of the gradient destination image.
If there are any values in the destination cliprect that do not contribute to the forward pooling result in the area specified by primaryOffset and sourceSize, due to large strides or dilation factors or simply because all forward pass induced values would be outside the source area, then those result values are set to zero.
The actual value of d out(x) / d in(y) depends on the pooling operation and these are defined in the subclasses of MPSCNNPoolingGradient.
NSObject.Function_instanceMethodForSelector_ret, NSObject.Function_methodForSelector_ret| Modifier | Constructor and Description |
|---|---|
protected |
MPSCNNPoolingGradient(org.moe.natj.general.Pointer peer) |
| Modifier and Type | Method and Description |
|---|---|
boolean |
_supportsSecureCoding()
This property must return YES on all classes that allow secure coding.
|
static boolean |
accessInstanceVariablesDirectly() |
static MPSCNNPoolingGradient |
alloc() |
static MPSCNNPoolingGradient |
allocWithZone(org.moe.natj.general.ptr.VoidPtr zone) |
static boolean |
automaticallyNotifiesObserversForKey(java.lang.String key) |
static void |
cancelPreviousPerformRequestsWithTarget(java.lang.Object aTarget) |
static void |
cancelPreviousPerformRequestsWithTargetSelectorObject(java.lang.Object aTarget,
org.moe.natj.objc.SEL aSelector,
java.lang.Object anArgument) |
static NSArray<java.lang.String> |
classFallbacksForKeyedArchiver() |
static org.moe.natj.objc.Class |
classForKeyedUnarchiver() |
static java.lang.String |
debugDescription_static() |
static java.lang.String |
description_static() |
static long |
hash_static() |
MPSCNNPoolingGradient |
init() |
MPSCNNPoolingGradient |
initWithCoder(NSCoder aDecoder)
NS_DESIGNATED_INITIALIZER
|
MPSCNNPoolingGradient |
initWithCoderDevice(NSCoder aDecoder,
java.lang.Object device)
NSSecureCoding compatability
|
MPSCNNPoolingGradient |
initWithDevice(java.lang.Object device)
Standard init with default properties per filter type
|
MPSCNNPoolingGradient |
initWithDeviceKernelWidthKernelHeight(MTLDevice device,
long kernelWidth,
long kernelHeight)
Initialize a gradient pooling filter
|
MPSCNNPoolingGradient |
initWithDeviceKernelWidthKernelHeightStrideInPixelsXStrideInPixelsY(MTLDevice device,
long kernelWidth,
long kernelHeight,
long strideInPixelsX,
long strideInPixelsY)
Initialize a gradient pooling filter
|
static NSObject.Function_instanceMethodForSelector_ret |
instanceMethodForSelector(org.moe.natj.objc.SEL aSelector) |
static NSMethodSignature |
instanceMethodSignatureForSelector(org.moe.natj.objc.SEL aSelector) |
static boolean |
instancesRespondToSelector(org.moe.natj.objc.SEL aSelector) |
static boolean |
isSubclassOfClass(org.moe.natj.objc.Class aClass) |
static NSSet<java.lang.String> |
keyPathsForValuesAffectingValueForKey(java.lang.String key) |
static MPSCNNPoolingGradient |
new_objc() |
static boolean |
resolveClassMethod(org.moe.natj.objc.SEL sel) |
static boolean |
resolveInstanceMethod(org.moe.natj.objc.SEL sel) |
void |
setSourceSize(MTLSize value)
[@property] sourceSize
|
static void |
setVersion_static(long aVersion) |
MTLSize |
sourceSize()
[@property] sourceSize
|
static org.moe.natj.objc.Class |
superclass_static() |
static boolean |
supportsSecureCoding() |
static long |
version_static() |
encodeToCommandBufferSourceGradientSourceImageGradientState, encodeToCommandBufferSourceGradientSourceImageGradientStateDestinationGradient, kernelOffsetX, kernelOffsetY, setKernelOffsetX, setKernelOffsetYappendBatchBarrier, clipRect, destinationFeatureChannelOffset, destinationImageAllocator, destinationImageDescriptorForSourceImagesSourceStates, encodeToCommandBufferPrimaryImageSecondaryImage, encodeToCommandBufferPrimaryImageSecondaryImageDestinationImage, encodeToCommandBufferPrimaryImageSecondaryImageDestinationStateDestinationStateIsTemporary, encodingStorageSizeForPrimaryImageSecondaryImageSourceStatesDestinationImage, isBackwards, isResultStateReusedAcrossBatch, isStateModified, padding, primaryDilationRateX, primaryDilationRateY, primaryEdgeMode, primaryKernelHeight, primaryKernelWidth, primaryOffset, primarySourceFeatureChannelMaxCount, primarySourceFeatureChannelOffset, primaryStrideInPixelsX, primaryStrideInPixelsY, resultStateForPrimaryImageSecondaryImageSourceStatesDestinationImage, secondaryDilationRateX, secondaryDilationRateY, secondaryEdgeMode, secondaryKernelHeight, secondaryKernelWidth, secondaryOffset, secondarySourceFeatureChannelMaxCount, secondarySourceFeatureChannelOffset, secondaryStrideInPixelsX, secondaryStrideInPixelsY, setClipRect, setDestinationFeatureChannelOffset, setDestinationImageAllocator, setPadding, setPrimaryEdgeMode, setPrimaryOffset, setPrimarySourceFeatureChannelMaxCount, setPrimarySourceFeatureChannelOffset, setPrimaryStrideInPixelsX, setPrimaryStrideInPixelsY, setSecondaryEdgeMode, setSecondaryOffset, setSecondarySourceFeatureChannelMaxCount, setSecondarySourceFeatureChannelOffset, setSecondaryStrideInPixelsX, setSecondaryStrideInPixelsY, temporaryResultStateForCommandBufferPrimaryImageSecondaryImageSourceStatesDestinationImagecopyWithZone, copyWithZoneDevice, device, encodeWithCoder, label, options, setLabel, setOptionsaccessibilityActivate, accessibilityActivationPoint, accessibilityAssistiveTechnologyFocusedIdentifiers, accessibilityAttributedHint, accessibilityAttributedLabel, accessibilityAttributedUserInputLabels, accessibilityAttributedValue, accessibilityContainerType, accessibilityCustomActions, accessibilityCustomRotors, accessibilityDecrement, accessibilityDragSourceDescriptors, accessibilityDropPointDescriptors, accessibilityElementAtIndex, accessibilityElementCount, accessibilityElementDidBecomeFocused, accessibilityElementDidLoseFocus, accessibilityElementIsFocused, accessibilityElements, accessibilityElementsHidden, accessibilityFrame, accessibilityHint, accessibilityIncrement, accessibilityLabel, accessibilityLanguage, accessibilityNavigationStyle, accessibilityPath, accessibilityPerformEscape, accessibilityPerformMagicTap, accessibilityRespondsToUserInteraction, accessibilityScroll, accessibilityTextualContext, accessibilityTraits, accessibilityUserInputLabels, accessibilityValue, accessibilityViewIsModal, addObserverForKeyPathOptionsContext, attemptRecoveryFromErrorOptionIndex, attemptRecoveryFromErrorOptionIndexDelegateDidRecoverSelectorContextInfo, autoContentAccessingProxy, awakeAfterUsingCoder, awakeFromNib, class_objc, classForCoder, classForKeyedArchiver, copy, dealloc, debugDescription, description, dictionaryWithValuesForKeys, didChangeValueForKey, didChangeValueForKeyWithSetMutationUsingObjects, didChangeValuesAtIndexesForKey, doesNotRecognizeSelector, fileManagerShouldProceedAfterError, fileManagerWillProcessPath, finalize_objc, forwardingTargetForSelector, forwardInvocation, hash, indexOfAccessibilityElement, isAccessibilityElement, isEqual, isKindOfClass, isMemberOfClass, isProxy, methodForSelector, methodSignatureForSelector, mutableArrayValueForKey, mutableArrayValueForKeyPath, mutableCopy, mutableOrderedSetValueForKey, mutableOrderedSetValueForKeyPath, mutableSetValueForKey, mutableSetValueForKeyPath, observationInfo, observeValueForKeyPathOfObjectChangeContext, performSelector, performSelectorInBackgroundWithObject, performSelectorOnMainThreadWithObjectWaitUntilDone, performSelectorOnMainThreadWithObjectWaitUntilDoneModes, performSelectorOnThreadWithObjectWaitUntilDone, performSelectorOnThreadWithObjectWaitUntilDoneModes, performSelectorWithObject, performSelectorWithObjectAfterDelay, performSelectorWithObjectAfterDelayInModes, performSelectorWithObjectWithObject, prepareForInterfaceBuilder, provideImageDataBytesPerRowOrigin_Size_UserInfo, removeObserverForKeyPath, removeObserverForKeyPathContext, replacementObjectForCoder, replacementObjectForKeyedArchiver, respondsToSelector, self, setAccessibilityActivationPoint, setAccessibilityAttributedHint, setAccessibilityAttributedLabel, setAccessibilityAttributedUserInputLabels, setAccessibilityAttributedValue, setAccessibilityContainerType, setAccessibilityCustomActions, setAccessibilityCustomRotors, setAccessibilityDragSourceDescriptors, setAccessibilityDropPointDescriptors, setAccessibilityElements, setAccessibilityElementsHidden, setAccessibilityFrame, setAccessibilityHint, setAccessibilityLabel, setAccessibilityLanguage, setAccessibilityNavigationStyle, setAccessibilityPath, setAccessibilityRespondsToUserInteraction, setAccessibilityTextualContext, setAccessibilityTraits, setAccessibilityUserInputLabels, setAccessibilityValue, setAccessibilityViewIsModal, setIsAccessibilityElement, setNilValueForKey, setObservationInfo, setShouldGroupAccessibilityChildren, setValueForKey, setValueForKeyPath, setValueForUndefinedKey, setValuesForKeysWithDictionary, shouldGroupAccessibilityChildren, superclass, validateValueForKeyError, validateValueForKeyPathError, valueForKey, valueForKeyPath, valueForUndefinedKey, willChangeValueForKey, willChangeValueForKeyWithSetMutationUsingObjects, willChangeValuesAtIndexesForKeyprotected MPSCNNPoolingGradient(org.moe.natj.general.Pointer peer)
public static boolean accessInstanceVariablesDirectly()
public static MPSCNNPoolingGradient alloc()
public static MPSCNNPoolingGradient allocWithZone(org.moe.natj.general.ptr.VoidPtr zone)
public static boolean automaticallyNotifiesObserversForKey(java.lang.String key)
public static void cancelPreviousPerformRequestsWithTarget(java.lang.Object aTarget)
public static void cancelPreviousPerformRequestsWithTargetSelectorObject(java.lang.Object aTarget,
org.moe.natj.objc.SEL aSelector,
java.lang.Object anArgument)
public static NSArray<java.lang.String> classFallbacksForKeyedArchiver()
public static org.moe.natj.objc.Class classForKeyedUnarchiver()
public static java.lang.String debugDescription_static()
public static java.lang.String description_static()
public static long hash_static()
public MPSCNNPoolingGradient init()
init in class MPSCNNGradientKernelpublic MPSCNNPoolingGradient initWithCoder(NSCoder aDecoder)
NSCodinginitWithCoder in interface NSCodinginitWithCoder in class MPSCNNGradientKernelpublic MPSCNNPoolingGradient initWithCoderDevice(NSCoder aDecoder, java.lang.Object device)
See @ref MPSKernel#initWithCoder.
initWithCoderDevice in class MPSCNNGradientKernelaDecoder - The NSCoder subclass with your serialized MPSCNNPoolingGradientdevice - The MTLDevice on which to make the MPSCNNPoolingGradientpublic MPSCNNPoolingGradient initWithDevice(java.lang.Object device)
MPSCNNGradientKernelinitWithDevice in class MPSCNNGradientKerneldevice - The device that the filter will be used on. May not be NULL.public MPSCNNPoolingGradient initWithDeviceKernelWidthKernelHeight(MTLDevice device, long kernelWidth, long kernelHeight)
device - The device the filter will run onkernelWidth - The width of the kernel. Can be an odd or even value.kernelHeight - The height of the kernel. Can be an odd or even value.public MPSCNNPoolingGradient initWithDeviceKernelWidthKernelHeightStrideInPixelsXStrideInPixelsY(MTLDevice device, long kernelWidth, long kernelHeight, long strideInPixelsX, long strideInPixelsY)
device - The device the filter will run onkernelWidth - The width of the kernel. Can be an odd or even value.kernelHeight - The height of the kernel. Can be an odd or even value.strideInPixelsX - The input stride (upsampling factor) in the x dimension.strideInPixelsY - The input stride (upsampling factor) in the y dimension.public static NSObject.Function_instanceMethodForSelector_ret instanceMethodForSelector(org.moe.natj.objc.SEL aSelector)
public static NSMethodSignature instanceMethodSignatureForSelector(org.moe.natj.objc.SEL aSelector)
public static boolean instancesRespondToSelector(org.moe.natj.objc.SEL aSelector)
public static boolean isSubclassOfClass(org.moe.natj.objc.Class aClass)
public static NSSet<java.lang.String> keyPathsForValuesAffectingValueForKey(java.lang.String key)
public static MPSCNNPoolingGradient new_objc()
public static boolean resolveClassMethod(org.moe.natj.objc.SEL sel)
public static boolean resolveInstanceMethod(org.moe.natj.objc.SEL sel)
public void setSourceSize(MTLSize value)
An optional source size which defines together with primaryOffset, the set of input gradient pixels to take into account in the gradient computations.
A MTLSize that together with primaryOffset indicates which part of the source gradient to consider. If the area does not lie completely within the primary source image, the intersection between source area rectangle and primary source bounds is used. Default: A size where every component is NSUIntegerMax indicating the entire rest of the image, starting from an offset (see primaryOffset).
public static void setVersion_static(long aVersion)
public MTLSize sourceSize()
An optional source size which defines together with primaryOffset, the set of input gradient pixels to take into account in the gradient computations.
A MTLSize that together with primaryOffset indicates which part of the source gradient to consider. If the area does not lie completely within the primary source image, the intersection between source area rectangle and primary source bounds is used. Default: A size where every component is NSUIntegerMax indicating the entire rest of the image, starting from an offset (see primaryOffset).
public static org.moe.natj.objc.Class superclass_static()
public static boolean supportsSecureCoding()
public boolean _supportsSecureCoding()
NSSecureCoding_supportsSecureCoding in interface NSSecureCoding_supportsSecureCoding in class MPSCNNGradientKernelpublic static long version_static()