I'll cut to the chase, so let's get right to the code~
model = Model(inputs=[src, tgt], outputs=[y, flow]) # Inputs and outputs are given when defining a network (optimizer=Adam(lr=lr), loss=[ losses.cc3D(), ('l2')], loss_weights=[1.0, reg_param]) # Specify the loss when training the network, in case of multi-loss. loss weightscorresponds to each of the precedinglossweights,final outputlosscounterfeit medicines train_loss = model.train_on_batch( [X, atlas_vol], [atlas_vol, zero_flow]) # Start training, the correspondence between y_pred and y_true in loss is: # Output y with atlas_vol to count cc3Dloss, output flow with zero_flow to count gradientloss
Additional knowledge:Code run by keras server with fit_generator, loss,acc graph saving
I'll cut to the chase, so let's get right to the code~
import as plt ... //Data processing code Omit history = model.fit_generator( image_generator, steps_per_epoch=2000 // 32 , epochs=16, verbose=1, validation_data=image_generator_TEST, validation_steps=20 ) print(()) plt.switch_backend('agg') #Saving images on the server requires this to be set. //acc (['acc']) (['val_acc']) ('model accuracy') ('accuracy') ('epoch') (['train', 'test'], loc='upper left') ('') //loss (['loss']) (['val_loss']) ('loss') ('epoch') (['train', 'test'], loc='upper left') ('')
Above this finally figured out the correspondence in Keras multiloss introduction is all that I have shared with you, I hope to give you a reference, and I hope you support me more.