Python ImageDataGenerator.standardize方法代码示例

# 需要导入模块: from keras.preprocessing.image import ImageDataGenerator [as 别名]
# 或者: from keras.preprocessing.image.ImageDataGenerator import standardize [as 别名]

#.........这里部分代码省略.........
        else:
            hsv_augmentation = None

        # This will do preprocessing and realtime data augmentation:
        datagen = ImageDataGenerator(
            # set input mean to 0 over the dataset
            featurewise_center=da['featurewise_center'],
            # set each sample mean to 0
            samplewise_center=da['samplewise_center'],
            # divide inputs by std of the dataset
            featurewise_std_normalization=False,
            # divide each input by its std
            samplewise_std_normalization=da['samplewise_std_normalization'],
            zca_whitening=da['zca_whitening'],
            # randomly rotate images in the range (degrees, 0 to 180)
            rotation_range=da['rotation_range'],
            # randomly shift images horizontally (fraction of total width)
            width_shift_range=da['width_shift_range'],
            # randomly shift images vertically (fraction of total height)
            height_shift_range=da['height_shift_range'],
            horizontal_flip=da['horizontal_flip'],
            vertical_flip=da['vertical_flip'],
            hsv_augmentation=hsv_augmentation,
            zoom_range=da['zoom_range'],
            shear_range=da['shear_range'],
            channel_shift_range=da['channel_shift_range'])

        # Compute quantities required for featurewise normalization
        # (std, mean, and principal components if ZCA whitening is applied).
        datagen.fit(X_train, seed=0)

        # Apply normalization to test data
        for i in range(len(X_test)):
            X_test[i] = datagen.standardize(X_test[i])

        # Fit the model on the batches generated by datagen.flow().
        steps_per_epoch = X_train.shape[0] // batch_size
        model.save(model_chk_path.format(epoch=0).replace('.00.',
                                                          '.00.a.'))
        t0 = time.time()
        model.fit_generator(datagen.flow(X_train, Y_train,
                            batch_size=batch_size),
                            steps_per_epoch=steps_per_epoch,
                            epochs=nb_epoch,
                            validation_data=(X_test, Y_test),
                            callbacks=callbacks)
        t1 = time.time()
        # Train one epoch without augmentation to make sure data distribution
        # is fit well
        loss_history = history_cb.history["loss"]
        epochs_augmented_training = len(loss_history)
        model.fit(X_train, Y_train,
                  batch_size=batch_size,
                  epochs=nb_epoch,
                  validation_data=(X_test, Y_test),
                  shuffle=True,
                  callbacks=callbacks,
                  initial_epoch=len(loss_history))
        t2 = time.time()
    loss_history = history_cb.history["loss"]
    acc_history = history_cb.history["acc"]
    val_acc_history = history_cb.history["val_acc"]
    np_loss_history = np.array(loss_history)
    np_acc_history = np.array(acc_history)
    np_val_acc_history = np.array(val_acc_history)
    history_data = zip(list(range(1, len(np_loss_history) + 1)),

# 需要导入模块: from keras.preprocessing.image import ImageDataGenerator [as 别名]
# 或者: from keras.preprocessing.image.ImageDataGenerator import standardize [as 别名]

#Reading Testing Images
log.info('Loading Testing images.')	
test_images_full_info = []
for i in range(N_TRAIN+1, N_TRAIN+N_TRAIN_TEST+1):
	#print '\rLoading:',i-(N_TRAIN),'/',N_TRAIN_TEST,'.',
	test_images_full_info.append(load_imgs('train/' + str(i) + '.png'))
print '\n'

#Preprossesing Test Images
log.info('Preprosseing Test images.')
ZZ = np.array(test_images_full_info)
Z = np.zeros((len(ZZ),1,32,32))
for i in range(len(ZZ)):
	Z[i] = datagen.standardize(ZZ[i])
y_test = np.array(labels[N_TRAIN+1:N_TRAIN+1+N_TRAIN_TEST])
	
log.info('Starting Classifier.')
nn = Classifier(
    layers=[
	Convolution("Rectifier",channels=64,kernel_shape=(3,3)),
	Native(lasagne.Conv2DLayer, num_filters=32,filter_size=(3,3), nonlinearity=nl.leaky_rectify),
	Native(lasagne.MaxPool2DLayer, pool_size=(2,2)),
	Native(lasagne.DropoutLayer, p=0.25),
	Native(lasagne.Conv2DLayer, num_filters=64,filter_size=(3,3), nonlinearity=nl.leaky_rectify),
	Native(lasagne.Conv2DLayer, num_filters=64,filter_size=(3,3), nonlinearity=nl.leaky_rectify),
	Native(lasagne.MaxPool2DLayer, pool_size=(2,2)),


	#Convolution("Rectifier",channels=100,kernel_shape=(30,30)),