Enum cudnnBatchNormMode

cudnnBatchNormMode is an enumerated type used to specify the mode of operation in cudnnBatchNormalizationForwardInference(), cudnnBatchNormalizationForwardTraining(), cudnnBatchNormalizationBackward() and cudnnDeriveBNTensorDescriptor() routines.

Namespace: ManagedCuda.CudaDNN

Assembly: CudaDNN.dll

Syntax

public enum cudnnBatchNormMode

Fields

Name	Description
BatchNormPerActivation	Normalization is performed per-activation. This mode is intended to be used after nonconvolutional network layers. In this mode bnBias and bnScale tensor dimensions are 1xCxHxW.
BatchNormSpatial	Normalization is performed over N+spatial dimensions. This mode is intended for use after convolutional layers (where spatial invariance is desired). In this mode bnBias, bnScale tensor dimensions are 1xCx1x1.
BatchNormSpatialPersitent	bnScale, bnBias tensor dims are 1xCx1x1 (one value per C-dim normalized over Nx1xHxW subtensors). May be faster than CUDNN_BATCHNORM_SPATIAL but imposes some limits on the range of values