Convolutions

Conv1D

layer eddl::Conv1D(layer parent, int filters, vector<int> kernel_size, vector<int> strides = {1}, string padding = "same", bool use_bias = true, int groups = 1, const vector<int> dilation_rate = {1}, string name = "")

1D Convolution layer.

Parameters

  • parent – Parent layer

  • filters – Integer, the dimensionality of the output space (i.e. the number of output filters in the convolution)

  • kernel_size – Vector of 1 integers, specifying the height and width of the 2D convolution window

  • strides – Vector of 1 integers, specifying the strides of the convolution along the height and width

  • padding – One of “none”, “valid” or “same”

  • use_bias – Boolean, whether the layer uses a bias vector

  • groups – Number of blocked connections from input channels to output channels

  • dilation_rate – Vector of 1 integers, specifying the dilation rate to use for dilated convolution

  • name – A name for the operation

Returns

Convolution layer

Example:

l = Conv1D(l, 16, {3}, {1});

Conv2D

layer eddl::Conv2D(layer parent, int filters, const vector<int> &kernel_size, const vector<int> &strides = {1, 1}, string padding = "same", bool use_bias = true, int groups = 1, const vector<int> &dilation_rate = {1, 1}, string name = "")

2D Convolution layer.

Parameters

  • parent – Parent layer

  • filters – Integer, the dimensionality of the output space (i.e. the number of output filters in the convolution)

  • kernel_size – Vector of 2 integers, specifying the height and width of the 2D convolution window

  • strides – Vector of 2 integers, specifying the strides of the convolution along the height and width

  • padding – One of “none”, “valid” or “same”

  • use_bias – Boolean, whether the layer uses a bias vector

  • groups – Number of blocked connections from input channels to output channels

  • dilation_rate – Vector of 2 integers, specifying the dilation rate to use for dilated convolution

  • name – A name for the operation

Returns

Convolution layer

Example:

l = Conv2D(l, 32, {3,3}, {1,1});

Conv3D

layer eddl::Conv3D(layer parent, int filters, const vector<int> &kernel_size, const vector<int> &strides = {1, 1, 1}, string padding = "same", bool use_bias = true, int groups = 1, const vector<int> &dilation_rate = {1, 1, 1}, string name = "")

3D Convolution layer.

Parameters

  • parent – Parent layer

  • filters – Integer, the dimensionality of the output space (i.e. the number of output filters in the convolution)

  • kernel_size – Vector of 3 integers, specifying the depth, height and width of the 3D convolution window

  • strides – Vector of 3 integers, specifying the strides of the convolution along the depth, height and width

  • padding – One of “none”, “valid” or “same”

  • use_bias – Boolean, whether the layer uses a bias vector

  • groups – Number of blocked connections from input channels to output channels

  • dilation_rate – Vector of 3 integers, specifying the dilation rate to use for dilated convolution

  • name – A name for the operation

Returns

Convolution layer 

l = Conv3D(l, 32, {3, 3, 3}, {1, 1, 1}, "same");

Pointwise Convolution 2D

layer eddl::PointwiseConv2D(layer parent, int filters, const vector<int> &strides = {1, 1}, bool use_bias = true, int groups = 1, const vector<int> &dilation_rate = {1, 1}, string name = "")

Pointwise 2D convolution.

Parameters

  • parent – Parent layer

  • filters – Integer, the dimensionality of the output space (i.e. the number of output filters in the convolution)

  • strides – Vector of 2 integers, specifying the strides of the convolution along the height and width

  • use_bias – Boolean, whether the layer uses a bias vector

  • groups – Number of blocked connections from input channels to output channels

  • dilation_rate – Vector of 2 integers, specifying the dilation rate to use for dilated convolution

  • name – A name for the operation

Returns

Convolution layer

Example:

l = PointwiseConv2D(l, 32, {3,3}, {1,1});

Depthwise Convolution 2D

layer eddl::DepthwiseConv2D(layer parent, const vector<int> &kernel_size, const vector<int> &strides = {1, 1}, string padding = "same", bool use_bias = true, const vector<int> &dilation_rate = {1, 1}, string name = "")

DepthwiseConv 2D convolution.

Parameters

  • parent – Parent layer

  • kernel_size – The depth, height and width of the convolution window

  • strides – Vector of 2 integers, specifying the strides of the convolution along the height and width

  • use_bias – Boolean, whether the layer uses a bias vector

  • dilation_rate – Vector of 2 integers, specifying the dilation rate to use for dilated convolution

  • name – A name for the operation

Returns

Convolution layer

Example:

l = DepthwiseConv2D(l, {3,3}, {1,1});

2D UpSampling

Soon to be deprecated. We recommend the use of the Resize layer.

layer eddl::UpSampling2D(layer parent, const vector<int> &size, string interpolation = "nearest", string name = "")

2D Upsampling layer.

Identical to the scale transformation, it is an alias of the Resize layer.

Parameters

  • parent – Parent layer

  • size – Vector of 2 integers. The upsampling factors for rows and columns

  • interpolation – (Deprecated) A string, only “nearest” is valid

  • name – A name for the operation

Returns

Output layer after upsampling operation

Example:

l = UpSampling2D(l, {2, 2});

3D UpSampling

UpSampling for 3D images

layer eddl::UpSampling3D(layer parent, vector<int> new_shape, bool reshape = true, string da_mode = "constant", float constant = 0.0f, string coordinate_transformation_mode = "asymmetric", string name = "")

3D Upsampling layer. Similar to Resize but for 3D images

Parameters

  • parent – Parent layer

  • new_shape – Vector with layer/images desired new shape

  • reshape – If True, the output shape will be new_shape (classical scale; recommended). If False, the output shape will be the input shape (scale<100%: scale + padding; scale >100%: crop + scale)

  • da_mode – One of “nearest”, “constant”, (ToDo: “mirror”, “reflect”, “wrap”, “original”)

  • constant – Fill value for area outside the resized image, it is used for all channels respectively

  • coordinate_transformation_mode – This attribute describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor.

Returns

Output of scale transformation

Example:

l = UpSampling3D(l, {32, 32, 32});

2D Convolutional Transpose

layer eddl::ConvT2D(layer parent, int filters, const vector<int> &kernel_size, const vector<int> &strides = {1, 1}, string padding = "same", bool use_bias = true, int groups = 1, const vector<int> &dilation_rate = {1, 1}, string name = "")

2D Transposed convolution layer (sometimes called Deconvolution).

The need for transposed convolutions generally arises from the desire to use a transformation going in the opposite direction of a normal convolution, i.e., from something that has the shape of the output of some convolution to something that has the shape of its input while maintaining a connectivity pattern that is compatible with said convolution.

Parameters

  • parent – Parent layer

  • filters – The dimensionality of the output space (i.e. the number of output filters in the convolution)

  • kernel_size – The height and width of the 2D convolution window

  • padding – One of “valid” or “same”

  • dilation_rate – The dilation rate to use for dilated convolution. Spacing between kernel elements

  • strides – The strides of the convolution along the height and width

  • use_bias – Boolean, whether the layer uses a bias vector

  • name – A name for the operation

Returns

Output layer after upsampling operation 

l = ConvT2D(l, 32, {3, 3}, {1, 1}, "same");

3D Convolutional Transpose

layer eddl::ConvT3D(layer parent, int filters, const vector<int> &kernel_size, const vector<int> &strides = {1, 1, 1}, string padding = "same", bool use_bias = true, int groups = 1, const vector<int> &dilation_rate = {1, 1, 1}, string name = "")

3D Transposed convolution layer (sometimes called Deconvolution).

The need for transposed convolutions generally arises from the desire to use a transformation going in the opposite direction of a normal convolution, i.e., from something that has the shape of the output of some convolution to something that has the shape of its input while maintaining a connectivity pattern that is compatible with said convolution.

Parameters

  • parent – Parent layer

  • filters – The dimensionality of the output space (i.e. the number of output filters in the convolution)

  • kernel_size – The depth, height and width of the 3D convolution window

  • padding – One of “valid” or “same”

  • dilation_rate – The dilation rate to use for dilated convolution. Spacing between kernel elements

  • strides – The strides of the convolution along the depth, height and width

  • use_bias – Boolean, whether the layer uses a bias vector

  • name – A name for the operation

Returns

Output layer after upsampling operation 

l = ConvT3D(l, 32, {3, 3, 3}, {1, 1, 1}, "same");