niftynet.layer.channel_sparse_convolution module

class niftynet.layer.channel_sparse_convolution.ChannelSparseBNLayer(n_dense_channels, *args, **kwargs)

Bases: niftynet.layer.bn.BNLayer

Channel sparse convolutions perform convolulations over a subset of image channels and generate a subset of output channels. This enables spatial dropout without wasted computations

layer_op(inputs, is_training, mask, use_local_stats=False)

Parameters: inputs: image to normalize. This typically represents a sparse subset of channels

from a sparse convolution.

is_training: boolean that is True during training. When True, the layer uses batch

statistics for normalization and records a moving average of means and variances. When False, the layer uses previously computed moving averages for normalization

mask: 1-Tensor with a binary mask identifying the sparse channels represented in inputs

class niftynet.layer.channel_sparse_convolution.ChannelSparseConvLayer(*args, **kwargs)

Bases: niftynet.layer.convolution.ConvLayer

Channel sparse convolutions perform convolulations over a subset of image channels and generate a subset of output channels. This enables spatial dropout without wasted computations

layer_op(input_tensor, input_mask, output_mask)

Parameters: input_tensor: image to convolve with kernel input_mask: 1-Tensor with a binary mask of input channels to use

If this is None, all channels are used.

output_mask: 1-Tensor with a binary mask of output channels to generate

If this is None, all channels are used and the number of output channels is set at graph-creation time.

class niftynet.layer.channel_sparse_convolution.ChannelSparseConvolutionalLayer(n_output_chns, kernel_size=3, stride=1, dilation=1, padding='SAME', with_bias=False, with_bn=True, acti_func=None, w_initializer=None, w_regularizer=None, b_initializer=None, b_regularizer=None, moving_decay=0.9, eps=1e-05, name='conv')

Bases: niftynet.layer.base_layer.TrainableLayer

This class defines a composite layer with optional components:

channel sparse convolution -> batchwise-spatial dropout -> batch_norm -> activation

The b_initializer and b_regularizer are applied to the ChannelSparseConvLayer The w_initializer and w_regularizer are applied to the ChannelSparseConvLayer, the batch normalisation layer, and the activation layer (for ‘prelu’)

layer_op(input_tensor, input_mask=None, is_training=None, keep_prob=None)

class niftynet.layer.channel_sparse_convolution.ChannelSparseDeconvLayer(*args, **kwargs)

Bases: niftynet.layer.deconvolution.DeconvLayer

Channel sparse convolutions perform convolulations over a subset of image channels and generate a subset of output channels. This enables spatial dropout without wasted computations

layer_op(input_tensor, input_mask=None, output_mask=None)

Parameters: input_tensor: image to convolve with kernel input_mask: 1-Tensor with a binary mask of input channels to use

If this is None, all channels are used.

output_mask: 1-Tensor with a binary mask of output channels to generate

If this is None, all channels are used and the number of output channels is set at graph-creation time.