def model_DenseNet():
model_dense = RGCSA.ResneXt_IN((1, img_rows, img_cols, img_channels), cardinality=8, classes=9)
RMS = RMSprop(lr=0.0003)
# Let's train the model using RMSprop
def mycrossentropy(y_true, y_pred, e=0.1):
loss1 = K.categorical_crossentropy(y_true, y_pred)
loss2 = K.categorical_crossentropy(K.ones_like(y_pred) / nb_classes, y_pred) # K.ones_like(y_pred) / nb_classes
return (1 - e) * loss1 + e * loss2
model_dense.compile(loss=mycrossentropy, optimizer=RMS, metrics=['accuracy'])
return model_dense
def model_DenseNet():
model_dense = RGCSA.ResneXt_IN((1, img_rows, img_cols, img_channels), classes=16)
RMS = RMSprop(lr=0.0003)
def mycrossentropy(y_true, y_pred, e=0.1):
loss1 = K.categorical_crossentropy(y_true, y_pred)
loss2 = K.categorical_crossentropy(K.ones_like(y_pred) / nb_classes, y_pred) # K.ones_like(y_pred) / nb_classes
return (1 - e) * loss1 + e * loss2
model_dense.compile(loss=mycrossentropy, optimizer=RMS, metrics=['accuracy']) # categorical_crossentropy
model_dense.summary()
# plot_model(model_dense, show_shapes=True, to_file='./model_ResNeXt_GroupChannel_Space_Attention.png')
return model_dense