Package apple.metalperformanceshaders

Class MPSCNNSubPixelConvolutionDescriptor

  • All Implemented Interfaces:
    NSCoding, NSCopying, NSSecureCoding, NSObject
    public class MPSCNNSubPixelConvolutionDescriptor
extends MPSCNNConvolutionDescriptor
    MPSCNNSubPixelConvolutionDescriptor can be used to create MPSCNNConvolution object that does sub pixel upsamling and reshaping opeartion as described in http://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Shi_Real-Time_Single_Image_CVPR_2016_paper.pdf Conceptually MPSCNNConvolution with subPixelScaleFactor > 1 can be thought of as filter performing regular CNN convolution producing N output feature channels at each pixel of an intermediate MPSImage followed by a kernel that rearranges/reshapes these N channels at each pixel of intermediate MPSImage into a pixel block of size subPixelScaleFactor x subPixelScaleFactor with N/(subPixelScaleFactor * subPixelScaleFactor) featureChannels at each pixel of this pixel block. Thus each pixel in intermedaite MPSImage with N channels map to subPixelScaleFactor x subPixelScaleFactor pixel block in final destination MPSImage with N/(subPixelScaleFactor * subPixelScaleFactor) featureChannels. MPSCNNConvolution with subPixelScaleFactor > 1 fuses the convolution and reshaping operation into single compute kernel thus not only saving DRAM roundtrip but also memory needed for intermediate MPSImage had these operation done separately. Let N be the value of outputFeatureChannels property and let r = subPixelScaleFactor. Conceptually Convolution will produce intermedaite image Io of dimensions (treated as 3D tensor) width x height x N where width = (clipRect.size.width + r - 1) / r height = (clipRect.size.height + r -1) / r Reshaping happens as follows [@code] Destination[clipRect.origin.x+x][clipRect.origin.y+y][c] = Io[ floor(x/r) ][ floor(y/r) ][ (N/r^2) * ( r * mod(y,r) + mod(x,r) ) + c ] where x in [0,clipRect.size.width-1], y in [0,clipRect.size.height-1], c in [0,N/r^2 - 1] [@endcode] The following conditions must be met: 1) N (outputFeatureChannels) must be multiple of r^2 (subPixelScaleFactor * subPixelScaleFactor). 2) The destination MPSImage to encode call must have at least N/r^2 + destinationFeatureChannelOffset channels. 3) Number of feature channels in reshaped output image (N/r^2) can be any value when groups = 1 but must be multiple of 4 when groups > 1.

    • Constructor Detail

      • MPSCNNSubPixelConvolutionDescriptor

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

    • Method Detail

      • accessInstanceVariablesDirectly

        public static boolean accessInstanceVariablesDirectly()

      • allocWithZone

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

      • 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()

      • cnnConvolutionDescriptorWithKernelWidthKernelHeightInputFeatureChannelsOutputFeatureChannels

        public static java.lang.Object cnnConvolutionDescriptorWithKernelWidthKernelHeightInputFeatureChannelsOutputFeatureChannels​(long kernelWidth,
                                                                                                                            long kernelHeight,
                                                                                                                            long inputFeatureChannels,
                                                                                                                            long outputFeatureChannels)

      • cnnConvolutionDescriptorWithKernelWidthKernelHeightInputFeatureChannelsOutputFeatureChannelsNeuronFilter

        public static java.lang.Object cnnConvolutionDescriptorWithKernelWidthKernelHeightInputFeatureChannelsOutputFeatureChannelsNeuronFilter​(long kernelWidth,
                                                                                                                                        long kernelHeight,
                                                                                                                                        long inputFeatureChannels,
                                                                                                                                        long outputFeatureChannels,
                                                                                                                                        MPSCNNNeuron neuronFilter)

      • 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)

      • new_objc

        public static java.lang.Object new_objc()

      • resolveClassMethod

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

      • resolveInstanceMethod

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

      • setSubPixelScaleFactor

        public void setSubPixelScaleFactor​(long value)
        [@property] subPixelScaleFactor Upsampling scale factor. Each pixel in input is upsampled into a subPixelScaleFactor x subPixelScaleFactor pixel block by rearranging the outputFeatureChannels as described above. Default value is 1.

      • setVersion_static

        public static void setVersion_static​(long aVersion)

      • subPixelScaleFactor

        public long subPixelScaleFactor()
        [@property] subPixelScaleFactor Upsampling scale factor. Each pixel in input is upsampled into a subPixelScaleFactor x subPixelScaleFactor pixel block by rearranging the outputFeatureChannels as described above. Default value is 1.

      • superclass_static

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

      • supportsSecureCoding

        public static boolean supportsSecureCoding()

      • _supportsSecureCoding

        public boolean _supportsSecureCoding()
        Description copied from interface: NSSecureCoding
        This property must return YES on all classes that allow secure coding. Subclasses of classes that adopt NSSecureCoding and override initWithCoder: must also override this method and return YES. The Secure Coding Guide should be consulted when writing methods that decode data.
        Specified by:
        _supportsSecureCoding in interface NSSecureCoding
        Overrides:
        _supportsSecureCoding in class MPSCNNConvolutionDescriptor

      • version_static

        public static long version_static()