def Res2DModule_c(self, n_input, in_layer, res_flag_2D=None, size=3, name=None):
if name == None:
name = 'Res2D_down_'+str(self.module_count)+'_'
self.module_count += 1
if res_flag_2D is None:
res_flag_2D = self.res_flag_2D
self.conv_2D_layers.append(Conv2DLayer(
n_input_feat=n_input,
n_output_feat=n_input//2,
n_size=1,
in_layers=[in_layer, res_flag_2D, self.training_placeholder], name=name+'conv_b1'))
in_layer_branch1 = self.conv_2D_layers[-1]
self.conv_2D_layers.append(Conv2DLayer(
n_input_feat=n_input,
n_output_feat=n_input//2,
n_size=1,
in_layers=[in_layer, res_flag_2D, self.training_placeholder], name=name+'conv_a1'))
in_layer_branch2 = self.conv_2D_layers[-1]
self.conv_2D_layers.append(Conv2DLayer(
n_input_feat=n_input//2,
n_output_feat=n_input//2,
n_size=size,
in_layers=[in_layer_branch2, res_flag_2D, self.training_placeholder], name=name+'conv_a2'))
in_layer_branch2 = self.conv_2D_layers[-1]
self.conv_2D_layers.append(Conv2DLayer(
n_input_feat=n_input//2,
n_output_feat=n_input//2,
n_size=size,
in_layers=[in_layer_branch2, res_flag_2D, self.training_placeholder], name=name+'conv_a3'))
in_layer_branch2 = self.conv_2D_layers[-1]
self.conv_2D_layers.append(Conv2DLayer(
n_input_feat=n_input//2,
n_output_feat=n_input//2,
n_size=1,
in_layers=[in_layer_branch2, res_flag_2D, self.training_placeholder], name=name+'conv_a4'))
in_layer_branch2 = self.conv_2D_layers[-1]
n_output = n_input//2
self.res_layers.append(ResAdd(in_layers=[in_layer_branch1, in_layer_branch2], name=name+'res_add'))
out_layer = self.res_layers[-1]
return n_output, out_layer
def Conv2DModule(self, n_input, in_layer):
self.conv_2D_layers = []
self.conv_2D_layers.append(
Conv2DLayer_RaptorX(n_input_feat=n_input,
n_output_feat=60,
n_size=3,
in_layers=[
in_layer, self.res_flag_2D,
self.training_placeholder
],
name='Res2D_global'))
in_layer = self.conv_2D_layers[-1]
n_input = 60
res_in = in_layer
for i in range(30):
self.conv_2D_layers.append(
Conv2DLayer_RaptorX(n_input_feat=n_input,
n_output_feat=60,
n_size=3,
in_layers=[
in_layer, self.res_flag_2D,
self.training_placeholder
],
name='Res2D_' + str(i) + '_conv_1'))
in_layer = self.conv_2D_layers[-1]
n_input = 60
self.conv_2D_layers.append(
Conv2DLayer_RaptorX(n_input_feat=n_input,
n_output_feat=60,
n_size=3,
in_layers=[
in_layer, self.res_flag_2D,
self.training_placeholder
],
name='Res2D_' + str(i) + '_conv_2'))
in_layer = self.conv_2D_layers[-1]
n_input = 60
self.res_layers.append(
ResAdd(in_layers=[in_layer, res_in],
name='Res2D_' + str(i) + '_res_add'))
res_in = self.res_layers[-1]
in_layer = self.res_layers[-1]
return n_input, in_layer
def Res1DModule_a(self, n_input, in_layer, size=3, name=None):
if name == None:
name = 'Res1D_up_'+str(self.module_count)+'_'
self.module_count += 1
self.conv_1D_layers.append(Conv1DLayer(
n_input_feat=n_input,
n_output_feat=n_input*2,
n_size=1,
in_layers=[in_layer, self.res_flag_1D, self.training_placeholder], name=name+'conv_b1'))
in_layer_branch1 = self.conv_1D_layers[-1]
self.conv_1D_layers.append(Conv1DLayer(
n_input_feat=n_input,
n_output_feat=n_input//2,
n_size=1,
in_layers=[in_layer, self.res_flag_1D, self.training_placeholder], name=name+'conv_a1'))
in_layer_branch2 = self.conv_1D_layers[-1]
self.conv_1D_layers.append(Conv1DLayer(
n_input_feat=n_input//2,
n_output_feat=n_input//2,
n_size=size,
in_layers=[in_layer_branch2, self.res_flag_1D, self.training_placeholder], name=name+'conv_a2'))
in_layer_branch2 = self.conv_1D_layers[-1]
self.conv_1D_layers.append(Conv1DLayer(
n_input_feat=n_input//2,
n_output_feat=n_input//2,
n_size=size,
in_layers=[in_layer_branch2, self.res_flag_1D, self.training_placeholder], name=name+'conv_a3'))
in_layer_branch2 = self.conv_1D_layers[-1]
self.conv_1D_layers.append(Conv1DLayer(
n_input_feat=n_input//2,
n_output_feat=n_input*2,
n_size=1,
in_layers=[in_layer_branch2, self.res_flag_1D, self.training_placeholder], name=name+'conv_a4'))
in_layer_branch2 = self.conv_1D_layers[-1]
n_output = n_input * 2
self.res_layers.append(ResAdd(in_layers=[in_layer_branch1, in_layer_branch2], name=name+'res_add'))
out_layer = self.res_layers[-1]
return n_output, out_layer