Skip to content

D
dlib
Commit fa2f35fe authored by Juha Reunanen's avatar Juha Reunanen Committed by Davis E. King
Browse files
Options

cont layer: allow to set the number of filters at runtime (#845) 

* cont layer: allow to set the number of filters at runtime
(implementation essentially copied from con layer)

* Document the new constructor cont_(num_con_outputs o), and the function set_num_filters
parent 6283a4b5
Hide whitespace changes
Inline Side-by-side
Showing with 55 additions and 18 deletions
dlib/dnn/layers.h
View file @ fa2f35fe
......@@ -164,7 +164,6 @@ namespace dlib
// set the initial bias values to zero
biases(params,filters.size()) = 0;
}
template <typename SUBNET>
......@@ -346,17 +345,22 @@ namespace dlib
static_assert(0 <= _padding_x && _padding_x < _nc, "The padding must be smaller than the filter size.");
cont_(
num_con_outputs o
) :
learning_rate_multiplier(1),
weight_decay_multiplier(1),
bias_learning_rate_multiplier(1),
bias_weight_decay_multiplier(0),
num_filters_(o.num_outputs),
padding_y_(_padding_y),
padding_x_(_padding_x)
{
DLIB_CASSERT(num_filters_ > 0);
}
long num_filters() const { return _num_filters; }
cont_() : cont_(num_con_outputs(_num_filters)) {}
long num_filters() const { return num_filters_; }
long nr() const { return _nr; }
long nc() const { return _nc; }
long stride_y() const { return _stride_y; }
......@@ -364,6 +368,14 @@ namespace dlib
long padding_y() const { return padding_y_; }
long padding_x() const { return padding_x_; }
void set_num_filters(long num)
{
DLIB_CASSERT(num > 0);
DLIB_CASSERT(get_layer_params().size() == 0,
"You can't change the number of filters in cont_ if the parameter tensor has already been allocated.");
num_filters_ = num;
}
double get_learning_rate_multiplier () const { return learning_rate_multiplier; }
double get_weight_decay_multiplier () const { return weight_decay_multiplier; }
void set_learning_rate_multiplier(double val) { learning_rate_multiplier = val; }
......@@ -402,6 +414,7 @@ namespace dlib
weight_decay_multiplier(item.weight_decay_multiplier),
bias_learning_rate_multiplier(item.bias_learning_rate_multiplier),
bias_weight_decay_multiplier(item.bias_weight_decay_multiplier),
num_filters_(item.num_filters_),
padding_y_(item.padding_y_),
padding_x_(item.padding_x_)
{
......@@ -427,6 +440,7 @@ namespace dlib
weight_decay_multiplier = item.weight_decay_multiplier;
bias_learning_rate_multiplier = item.bias_learning_rate_multiplier;
bias_weight_decay_multiplier = item.bias_weight_decay_multiplier;
num_filters_ = item.num_filters_;
return *this;
}
......@@ -434,18 +448,18 @@ namespace dlib
void setup (const SUBNET& sub)
{
long num_inputs = _nr*_nc*sub.get_output().k();
long num_outputs = _num_filters;
long num_outputs = num_filters_;
// allocate params for the filters and also for the filter bias values.
params.set_size(num_inputs*_num_filters + _num_filters);
params.set_size(num_inputs*num_filters_ + num_filters_);
dlib::rand rnd(std::rand());
randomize_parameters(params, num_inputs+num_outputs, rnd);
filters = alias_tensor(sub.get_output().k(),_num_filters, _nr, _nc);
biases = alias_tensor(1,_num_filters);
filters = alias_tensor(sub.get_output().k(), num_filters_, _nr, _nc);
biases = alias_tensor(1,num_filters_);
// set the initial bias values to zero
biases(params,_num_filters) = 0;
biases(params,filters.size()) = 0;
}
template <typename SUBNET>
......@@ -484,7 +498,7 @@ namespace dlib
{
serialize("cont_1", out);
serialize(item.params, out);
serialize(_num_filters, out);
serialize(item.num_filters_, out);
serialize(_nr, out);
serialize(_nc, out);
serialize(_stride_y, out);
......@@ -503,7 +517,6 @@ namespace dlib
{
std::string version;
deserialize(version, in);
long num_filters;
long nr;
long nc;
int stride_y;
......@@ -511,7 +524,7 @@ namespace dlib
if (version == "cont_1")
{
deserialize(item.params, in);
deserialize(num_filters, in);
deserialize(item.num_filters_, in);
deserialize(nr, in);
deserialize(nc, in);
deserialize(stride_y, in);
......@@ -526,14 +539,6 @@ namespace dlib
deserialize(item.bias_weight_decay_multiplier, in);
if (item.padding_y_ != _padding_y) throw serialization_error("Wrong padding_y found while deserializing dlib::con_");
if (item.padding_x_ != _padding_x) throw serialization_error("Wrong padding_x found while deserializing dlib::con_");
if (num_filters != _num_filters)
{
std::ostringstream sout;
sout << "Wrong num_filters found while deserializing dlib::con_" << std::endl;
sout << "expected " << _num_filters << " but found " << num_filters << std::endl;
throw serialization_error(sout.str());
}
if (nr != _nr) throw serialization_error("Wrong nr found while deserializing dlib::con_");
if (nc != _nc) throw serialization_error("Wrong nc found while deserializing dlib::con_");
if (stride_y != _stride_y) throw serialization_error("Wrong stride_y found while deserializing dlib::con_");
......@@ -592,6 +597,7 @@ namespace dlib
double weight_decay_multiplier;
double bias_learning_rate_multiplier;
double bias_weight_decay_multiplier;
long num_filters_;
int padding_y_;
int padding_x_;
......
dlib/dnn/layers_abstract.h
View file @ fa2f35fe
......@@ -941,6 +941,24 @@ namespace dlib
- #get_bias_weight_decay_multiplier() == 0
!*/
cont_(
num_con_outputs o
);
/*!
ensures
- #num_filters() == o.num_outputs
- #nr() == _nr
- #nc() == _nc
- #stride_y() == _stride_y
- #stride_x() == _stride_x
- #padding_y() == _padding_y
- #padding_x() == _padding_x
- #get_learning_rate_multiplier() == 1
- #get_weight_decay_multiplier() == 1
- #get_bias_learning_rate_multiplier() == 1
- #get_bias_weight_decay_multiplier() == 0
!*/
long num_filters(
) const;
/*!
......@@ -950,6 +968,19 @@ namespace dlib
of filters.
!*/
void set_num_filters(
long num
);
/*!
requires
- num > 0
- get_layer_params().size() == 0
(i.e. You can't change the number of filters in cont_ if the parameter
tensor has already been allocated.)
ensures
- #num_filters() == num
!*/
long nr(
) const;
/*!
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment