How to use Center_loss in fit_generator not fit

[FlyEgle]

fit

history_softmax = resnet_model.fit_generator( train_generator, steps_per_epoch=123520//BATCH_SIZE, epochs=EPOCHS, validation_data=val_generator, validation_steps=7119//BATCH_SIZE, callbacks=[lr] )

fit_generator

history2 = model_centerloss.fit_generator(train_generator, [y_train, random_y_train], batch_size=batch_size, epochs=epochs, verbose=1, validation_data=(val_generator, [y_test, random_y_test]), callbacks=[lr, modelcheckpoint])

[shamangary]

You have to define your own data generator and make the input X have two parameters.

hist = model.fit_generator(generator=data_generator_centerloss(X=[x_train, y_train_a_class_value], Y=[y_train_a_class,y_train_a, random_y_train_a], batch_size=batch_size),
                                   steps_per_epoch=train_num // batch_size,
                                   validation_data=([x_test,y_test_a_class_value], [y_test_a_class,y_test_a, random_y_test_a]),
                                   epochs=nb_epochs, verbose=1,
                                   callbacks=callbacks)

[FlyEgle]

Can you have a look my py script? i am not good at myself define keras func

[shamangary]

An example function of fit_generator. It should fit the center loss data structure.

def data_generator_centerloss(X,Y,batch_size):
    X1 = X[0]
    X2 = X[1]
    Y1 = Y[0]
    Y2 = Y[1]
    Y3 = Y[2]
    while True:
        idxs = np.random.permutation(len(X1))
        X1 = X1[idxs] #images
        X2 = X2[idxs] #labels for center loss
        Y1 = Y1[idxs]
        Y2 = Y2[idxs]
        Y3 = Y3[idxs]
        p1,p2,q1,q2,q3 = [],[],[],[],[]
        for i in range(len(X1)):
            p1.append(X1[i])
            p2.append(X2[i])
            q1.append(Y1[i])
            q2.append(Y2[i])
            q3.append(Y3[i])
            if len(p1) == batch_size:
                yield [np.array(p1),np.array(p2)],[np.array(q1),np.array(q2),np.array(q3)]
                p1,p2,q1,q2,q3 = [],[],[],[],[]
        if p1:
            yield [np.array(p1),np.array(p2)],[np.array(q1),np.array(q2),np.array(q3)]
            p1,p2,q1,q2,q3 = [],[],[],[],[]

[FlyEgle]

train_generator = generator_from_directory(data_train_agu(), train_path, SIZE, BATCH_SIZE, 'categorical')

val data generator

val_generator = generator_from_directory(data_val_agu(), val_path, SIZE, BATCH_SIZE, 'categorical')

if isCenterloss: random_y_train = np.random.rand(123520, 1) # x_train random_y_test = np.random.rand(7119, 1) # x_test

history2 = model_centerloss.fit_generator(train_generator, [y_train, random_y_train], 
                                          batch_size=batch_size, epochs=epochs,
                     verbose=1, validation_data=(val_generator, [y_test, random_y_test]),
                     callbacks=[lr, modelcheckpoint])

so that mean, i should redefine train_generator and val_generator

[shamangary]

Yes. You have to redefine them by yourself. The above data generator example I give you should be enough.

[FlyEgle]

ok，i do it.Can i have your wechat ？May be another bug cause.

[FlyEgle]

this is right?but it have wrong in keras

[vsatyakumar]

@FlyEgle Did you get this to work?

[FlyEgle]

yes, i have work for a myself generator

[vsatyakumar]

Same here. Works wonders!

[Dicksonchin93]

@FlyEgle Will you be kind enough to share your code? Thanks!

[alyato]

hi @FlyEgle Does your code do well using the center_loss. But it seemly does't work my code, the accuracy is low. i don't know how to settle it.Do you check the #3 and give me some suggestions. Thanks.