def get_model(): model = Sequential() model.add(Convolution2D(16, 5, 3, 3)) model.add(Activation('relu')) model.add(Maxout(2)) model.add(Convolution2D(16, 16 / 2, 3, 3)) model.add(Activation('relu')) model.add(Maxout(2)) model.add(Convolution2D(24, 16 / 2, 3, 3)) model.add(Activation('relu')) model.add(Maxout(2)) incept0, incept0_chan = model_tools.get_inception( input_channel=24 / 2, nr_c0_conv_1x1=24, nr_c1_conv_1x1=8, nr_c1_conv_3x3=24, nr_c2_conv_1x1=4, nr_c2_conv_5x5=12, nr_c3_conv_1x1=24, return_output_channels=True ) model.add(incept0) model.add(Maxout(2)) incept1, incept1_chan = model_tools.get_inception( input_channel=incept0_chan / 2, nr_c0_conv_1x1=24, nr_c1_conv_1x1=8, nr_c1_conv_3x3=24, nr_c2_conv_1x1=4, nr_c2_conv_5x5=12, nr_c3_conv_1x1=24, return_output_channels=True ) model.add(incept1) model.add(Maxout(2)) model.add(Convolution2D(32, incept1_chan / 2, 3, 3)) model.add(Activation('relu')) model.add(Maxout(2)) model.add(Convolution2D(48, 32 / 2, 3, 3)) model.add(Activation('relu')) model.add(Maxout(2)) # 1x1 here model.add(Convolution2D(1, 48 / 2, 1, 1)) # a fully connected layer model.add(Activation('sigmoid')) return model
def get_model(): model = Sequential() model.add(Convolution2D(8, 1, 3, 3)) model.add(Activation('relu')) model.add(Convolution2D(16, 8, 3, 3)) model.add(Activation('relu')) model.add(Dropout(0.2)) incept0, incept0_chan = model_tools.get_inception( input_channel=16, nr_c0_conv_1x1=16, nr_c1_conv_1x1=8, nr_c1_conv_3x3=16, nr_c2_conv_1x1=4, nr_c2_conv_5x5=8, nr_c3_conv_1x1=16, return_output_channels=True ) model.add(incept0) model.add(Convolution2D(24, incept0_chan, 3, 3)) model.add(Activation('relu')) model.add(Dropout(0.2)) model.add(Convolution2D(32, 24, 3, 3)) model.add(Activation('relu')) model.add(Dropout(0.2)) model.add(Convolution2D(48, 32, 3, 3)) model.add(Activation('relu')) model.add(Dropout(0.2)) # 1x1 here model.add(Convolution2D(1, 48, 1, 1)) # a fully connected layer model.add(Activation('sigmoid')) return model
def get_model(): model = Sequential() model.add(Convolution2D(16, 1, 3, 3)) model.add(Activation("relu")) model.add(Maxout(2)) model.add(Convolution2D(16, 8, 3, 3)) model.add(Activation("relu")) model.add(Maxout(2)) # incept0, incept0_chan = model_tools.get_inception( # input_channel=16 / 2, # nr_c0_conv_1x1=16, # nr_c1_conv_1x1=8, nr_c1_conv_3x3=16, # nr_c2_conv_1x1=4, nr_c2_conv_5x5=8, # nr_c3_conv_1x1=16, # return_output_channels=True # ) # model.add(incept0) # model.add(Maxout(2)) # incept1, incept1_chan = model_tools.get_inception( # input_channel=incept0_chan / 2, # nr_c0_conv_1x1=16, # nr_c1_conv_1x1=8, nr_c1_conv_3x3=16, # nr_c2_conv_1x1=4, nr_c2_conv_5x5=8, # nr_c3_conv_1x1=16, # return_output_channels=True # ) # model.add(incept1) # model.add(Maxout(2)) # incept2, incept2_chan = model_tools.get_inception( # input_channel=incept1_chan / 2, # nr_c0_conv_1x1=16, # nr_c1_conv_1x1=8, nr_c1_conv_3x3=16, # nr_c2_conv_1x1=4, nr_c2_conv_5x5=8, # nr_c3_conv_1x1=16, # return_output_channels=True # ) # model.add(incept2) # model.add(Maxout(2)) # model.add(Convolution2D(24, incept2_chan / 2, 3, 3)) model.add(Convolution2D(24, 16 / 2, 3, 3)) model.add(Activation("relu")) model.add(Maxout(2)) incept3, incept3_chan = model_tools.get_inception( input_channel=24 / 2, nr_c0_conv_1x1=24, nr_c1_conv_1x1=8, nr_c1_conv_3x3=24, nr_c2_conv_1x1=4, nr_c2_conv_5x5=12, nr_c3_conv_1x1=24, return_output_channels=True, ) model.add(incept3) model.add(Maxout(2)) # incept4, incept4_chan = model_tools.get_inception( # input_channel=incept3_chan / 2, # nr_c0_conv_1x1=24, # nr_c1_conv_1x1=8, nr_c1_conv_3x3=24, # nr_c2_conv_1x1=4, nr_c2_conv_5x5=12, # nr_c3_conv_1x1=24, # return_output_channels=True # ) # model.add(incept4) # model.add(Maxout(2)) # incept5, incept5_chan = model_tools.get_inception( # input_channel=incept4_chan / 2, # nr_c0_conv_1x1=24, # nr_c1_conv_1x1=8, nr_c1_conv_3x3=24, # nr_c2_conv_1x1=4, nr_c2_conv_5x5=12, # nr_c3_conv_1x1=24, # return_output_channels=True # ) # model.add(incept5) # model.add(Maxout(2)) # model.add(Convolution2D(32, incept5_chan / 2, 3, 3)) model.add(Convolution2D(32, incept3_chan / 2, 3, 3)) model.add(Activation("relu")) model.add(Maxout(2)) model.add(Convolution2D(48, 32 / 2, 3, 3)) model.add(Activation("relu")) model.add(Maxout(2)) # 1x1 here model.add(Convolution2D(1, 48 / 2, 1, 1)) # a fully connected layer model.add(Activation("sigmoid")) return model