diff --git a/dlib/dnn/layers_abstract.h b/dlib/dnn/layers_abstract.h
index a2175db2332aef9232a00fc9d6722cb248cdaa44..5cb687c00ee71a6f65604c17e8c42b8b88a95e66 100644
--- a/dlib/dnn/layers_abstract.h
+++ b/dlib/dnn/layers_abstract.h
@@ -522,7 +522,189 @@ namespace dlib
// ----------------------------------------------------------------------------------------
-// TODO, add spec for max_pool_, bn_, and affine_ layers.
+ class affine_
+ {
+ /*!
+ WHAT THIS OBJECT REPRESENTS
+ This is an implementation of the EXAMPLE_LAYER_ interface defined above.
+ In particular, it applies a simple pointwise linear transformation to an
+ input tensor. You can think of it as having two parameter tensors, A and
+ B, that each have the same dimensionality as the input tensor (except their
+ num_samples() dimensions are 1). If the input tensor is called INPUT
+ then the output of this layer is simply:
+ A*INPUT+B
+ where all operations are performed element wise and each sample in the
+ INPUT tensor is processed separately.
+ !*/
+
+ public:
+
+ affine_(
+ );
+
+ template <typename SUBNET> void setup (const SUBNET& sub);
+ void forward_inplace(const tensor& input, tensor& output);
+ void backward_inplace(const tensor& computed_output, const tensor& gradient_input, tensor& data_grad, tensor& params_grad);
+ const tensor& get_layer_params() const;
+ tensor& get_layer_params();
+ /*!
+ These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ !*/
+ };
+
+ void serialize(const affine_& item, std::ostream& out);
+ void deserialize(affine_& item, std::istream& in);
+ /*!
+ provides serialization support
+ !*/
+
+ template <typename SUBNET>
+ using affine = add_layer<affine_, SUBNET>;
+
+// ----------------------------------------------------------------------------------------
+
+ class bn_
+ {
+ /*!
+ WHAT THIS OBJECT REPRESENTS
+ This is an implementation of the EXAMPLE_LAYER_ interface defined above.
+ In particular, it defines a batch normalization layer that implements the
+ method described in the paper:
+ Batch Normalization: Accelerating Deep Network Training by Reducing
+ Internal Covariate Shift by Sergey Ioffe and Christian Szegedy
+
+ In particular, this layer produces output tensors with the same
+ dimensionality as the input tensors, except that the mean and variances of
+ the elements have been standardized.
+ !*/
+
+ public:
+ bn_(
+ );
+
+ template <typename SUBNET> void setup (const SUBNET& sub);
+ template <typename SUBNET> void forward(const SUBNET& sub, resizable_tensor& output);
+ template <typename SUBNET> void backward(const tensor& gradient_input, SUBNET& sub, tensor& params_grad);
+ const tensor& get_layer_params() const;
+ tensor& get_layer_params();
+ /*!
+ These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ !*/
+ };
+
+ void serialize(const bn_& item, std::ostream& out);
+ void deserialize(bn_& item, std::istream& in);
+ /*!
+ provides serialization support
+ !*/
+
+ template <typename SUBNET>
+ using bn = add_layer<bn_, SUBNET>;
+
+// ----------------------------------------------------------------------------------------
+
+ class max_pool_
+ {
+ /*!
+ WHAT THIS OBJECT REPRESENTS
+ This is an implementation of the EXAMPLE_LAYER_ interface defined above.
+ In particular, it defines a max pooling layer that takes an input tensor
+ and downsamples it. It does this by sliding a window over the images in an
+ input tensor and outputting, for each channel, the maximum element within
+ the window.
+
+ To be precise, if we call the input tensor IN and the output tensor OUT,
+ then OUT is defined as follows:
+ - OUT.num_samples() == IN.num_samples()
+ - OUT.k() == IN.k()
+ - OUT.nr() == IN.nr()/stride_y()
+ - OUT.nc() == IN.nc()/stride_x()
+ - for all valid s, k, r, and c:
+ - image_plane(OUT,s,k)(r,c) == max(subm_clipped(image_plane(IN,s,k),
+ r*stride_y(),
+ c*stride_x(),
+ nr(),
+ nc()))
+ !*/
+
+ public:
+
+ max_pool_ (
+ );
+ /*!
+ ensures
+ - #nr() == 3
+ - #nc() == 3
+ - #stride_y() == 1
+ - #stride_x() == 1
+ !*/
+
+ max_pool_(
+ long nr_,
+ long nc_,
+ int stride_y_ = 1,
+ int stride_x_ = 1
+ );
+ /*!
+ ensures
+ - #nr() == nr_
+ - #nc() == nc_
+ - #stride_y() == stride_y_
+ - #stride_x() == stride_x_
+ !*/
+
+ long nr(
+ ) const;
+ /*!
+ ensures
+ - returns the number of rows in the max pooling window.
+ !*/
+
+ long nc(
+ ) const;
+ /*!
+ ensures
+ - returns the number of columns in the max pooling window.
+ !*/
+
+ long stride_y(
+ ) const;
+ /*!
+ ensures
+ - returns the vertical stride used when scanning the max pooling window
+ over an image. That is, each window will be moved stride_y() pixels down
+ at a time when it moves over the image.
+ !*/
+
+ long stride_x(
+ ) const;
+ /*!
+ ensures
+ - returns the horizontal stride used when scanning the max pooling window
+ over an image. That is, each window will be moved stride_x() pixels down
+ at a time when it moves over the image.
+ !*/
+
+ template <typename SUBNET> void setup (const SUBNET& sub);
+ template <typename SUBNET> void forward(const SUBNET& sub, resizable_tensor& output);
+ template <typename SUBNET> void backward(const tensor& computed_output, const tensor& gradient_input, SUBNET& sub, tensor& params_grad);
+ const tensor& get_layer_params() const;
+ tensor& get_layer_params();
+ /*!
+ These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ Note that this layer doesn't have any parameters, so the tensor returned by
+ get_layer_params() is always empty.
+ !*/
+ };
+
+ void serialize(const max_pool_& item, std::ostream& out);
+ void deserialize(max_pool_& item, std::istream& in);
+ /*!
+ provides serialization support
+ !*/
+
+ template <typename SUBNET>
+ using max_pool = add_layer<max_pool_, SUBNET>;
// ----------------------------------------------------------------------------------------
@@ -549,6 +731,8 @@ namespace dlib
tensor& get_layer_params();
/*!
These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ Note that this layer doesn't have any parameters, so the tensor returned by
+ get_layer_params() is always empty.
!*/
};
@@ -586,6 +770,8 @@ namespace dlib
tensor& get_layer_params();
/*!
These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ Note that this layer doesn't have any parameters, so the tensor returned by
+ get_layer_params() is always empty.
!*/
};
@@ -623,6 +809,8 @@ namespace dlib
tensor& get_layer_params();
/*!
These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ Note that this layer doesn't have any parameters, so the tensor returned by
+ get_layer_params() is always empty.
!*/
};
@@ -668,6 +856,8 @@ namespace dlib
tensor& get_layer_params();
/*!
These functions are implemented as described in the EXAMPLE_LAYER_ interface.
+ Note that this layer doesn't have any parameters, so the tensor returned by
+ get_layer_params() is always empty.
!*/
};