NAME

MPSCNNConvolutionNode

SYNOPSIS

#import <MPSNNGraphNodes.h>

Inherits MPSNNFilterNode.

Inherited by MPSCNNBinaryConvolutionNode, MPSCNNConvolutionTransposeNode, and MPSCNNFullyConnectedNode.

Instance Methods

(nonnull instancetype) - initWithSource:weights:

Class Methods

(nonnull instancetype) + nodeWithSource:weights:

Properties

MPSNNConvolutionAccumulatorPrecisionOption accumulatorPrecision
MPSCNNConvolutionGradientStateNode * convolutionGradientState

Method Documentation

- (nonnull instancetype) initWithSource: (MPSNNImageNode *__nonnull) sourceNode(nonnull id< MPSCNNConvolutionDataSource >) weights

Init a node representing a MPSCNNConvolution kernel

Parameters:

sourceNode The MPSNNImageNode representing the source MPSImage for the filter
weights A pointer to a valid object conforming to the MPSCNNConvolutionDataSource protocol. This object is provided by you to encapsulate storage for convolution weights and biases. If it is used for training, it may not have a neuron embedded in the convolution descriptor.

Returns:

A new MPSNNFilter node for a MPSCNNConvolution kernel.

Implemented in MPSCNNFullyConnectedNode.

+ (nonnull instancetype) nodeWithSource: (MPSNNImageNode *__nonnull) sourceNode(nonnull id< MPSCNNConvolutionDataSource >) weights

Init an autoreleased not representing a MPSCNNConvolution kernel

Parameters:

sourceNode The MPSNNImageNode representing the source MPSImage for the filter
weights A pointer to a valid object conforming to the MPSCNNConvolutionDataSource protocol. This object is provided by you to encapsulate storage for convolution weights and biases. If it is used for training, it may not have a neuron embedded in the convolution descriptor.

Returns:

A new MPSNNFilter node for a MPSCNNConvolution kernel.

Implemented in MPSCNNFullyConnectedNode.

Property Documentation

- (MPSNNConvolutionAccumulatorPrecisionOption) accumulatorPrecision [read], [write], [nonatomic], [assign]

Set the floating-point precision used by the convolution accumulator Default: MPSNNConvolutionAccumulatorPrecisionOptionFloat

- (MPSCNNConvolutionGradientStateNode*) convolutionGradientState [read], [nonatomic], [assign]

A node to represent a MPSCNNConvolutionGradientState object Use this if the convolution is mirrored by a convolution transpose node later on in the graph to make sure that the size of the image returned from the convolution transpose matches the size of the image passed in to this node.

Author

Generated automatically by Doxygen for MetalPerformanceShaders.framework from the source code.