Class Conv2DBackpropFilterFusion
Defined in File conv2d_backprop_filter_fusion.h
Inheritance Relationships
Base Type
public Conv2DBackpropFilter
Class Documentation
-
class Conv2DBackpropFilterFusion : public Conv2DBackpropFilter
Conv2DBackpropFilterFusion defined Conv2DBackpropFilter operator prototype of lite.
Public Functions
-
inline Conv2DBackpropFilterFusion()
Constructor.
-
void Init(const int64_t out_channel, const std::vector<int64_t> &kernel_size, const PadMode &pad_mode = VALID, const std::vector<int64_t> &pad_list = {0, 0, 0, 0}, const int64_t mode = 1, const std::vector<int64_t> &stride = {1, 1}, const std::vector<int64_t> &dilation = {1, 1, 1, 1}, const int64_t group = 1, const Format &format = NCHW, const ActivationType activation_type = NO_ACTIVATION)
Method to init the op’s attributes.
- Parameters
out_channel – [in] Define the number of output channel.
kernel_size – [in] Define the size of the filter kernel.
pad_mode – [in] Define the padding method.
pad_list – [in] Define the concrete padding value on H and W dimension.
mode – [in] Define the category of conv, which is useless on lite.
stride – [in] Define the moving size of the filter kernel.
dilation – [in] Define the coefficient of expansion of the filter kernel, which is useful for dilated convolution.
group – [in] Define the number of group.
format – [in] Define the format of input tensor.
activation_type – [in] Define the activation type.
-
void set_activation_type(const ActivationType activation_type)
Method to set activation type.
- Parameters
activation_type – [in] Define the activation type.
-
void set_in_channel(const int64_t in_channel)
Method to set in_channel attribute.
- Parameters
in_channel – [in] Define the number of input channel.
-
ActivationType get_activation_type() const
Method to get activation type.
- Returns
activation type.
-
int64_t get_in_channel() const
Method to get in_channel attribute.
- Returns
the number of input channel.
-
inline Conv2DBackpropFilterFusion()