Python Keras model.fit_发电机在第一个时期停止/冻结

我正在设置一个简单的u-net实现，将Gipl文件转换为PIL图像。不幸的是，在添加数据生成器以更有效地分配GPU性能后，网络在初始化第一个历元后停止工作，没有任何进一步的输出

当我禁用模型上的use_多重处理时，网络会为第一个历元生成图像，但很快就会耗尽内存。再次启用该选项后，不会生成图像，也不会生成输出。至少应该开始准备图像

unet模式：

model = Model(inputs=[inputs], outputs=[outputs])
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=[mean_iou])
model.summary()

params_train = {'dim': (512,512,1),
          'batch_size': 16,
          'n_classes': 2,
          'n_channels': 1,
          'shuffle': True}

X, Xv, Y, Yv = getSubsets(X_train_path, Y_train_path)

training_generator = DataGenerator(X, Y, **params_train)
validation_generator = validation_generator = DataGenerator(Xv, Yv, **params_train)  

earlystopper = EarlyStopping(patience=5, verbose=1)
checkpointer = ModelCheckpoint('model-2019-1.h5', verbose=1, save_best_only=True)# path of model
print('last output, no output of datagenerator')

model_checkpoint = ModelCheckpoint('unet_fmr.hdf5', monitor='loss',verbose=1, save_best_only=True)

results = model.fit_generator(generator=training_generator,
                    validation_data=validation_generator,
                    verbose=1,
                    use_multiprocessing = True,
                    epochs=30,
                    callbacks=[model_checkpoint])

已编辑的数据生成器：

class DataGenerator(keras.utils.Sequence):
    def __init__(self, list_IDs, labels, batch_size, dim, n_channels=1,
             n_classes=2, shuffle=True):
        #...initialize variables
        self.on_epoch_end()

    def __getitem__(self, index):
        'Generate one batch of data'
        indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]

        X, Y = self.__data_generation(indexes)
        return X, Y

    def __len__(self):
        return int(np.floor(len(self.list_IDs) / self.batch_size))

    def on_epoch_end(self):        
        self.indexes = np.arange(len(self.list_IDs))
        if self.shuffle == True:
            np.random.shuffle(self.indexes)

    def getImage(self, path):
        x_train_array = getGiplAsArray(path)
        x_train_i = getResizedGiplImageByArray(x_train_array)
        return x_train_i

    def make3D(self, img):
        img = np.reshape(img, (img.shape[0], img.shape[1], 1))
        return img

    def __data_generation(self, indexes):
        print('__data_generation()')
        list_IDs_temp = [self.list_IDs[k] for k in indexes]
        list_labels_temp = [self.labels[k] for k in indexes]

        X = np.empty((self.batch_size, *self.dim))
        Y = np.empty((self.batch_size, *self.dim))

        for i, ID in enumerate(list_IDs_temp):
            temp = self.getImage(ID)
            X[i,] = self.make3D(temp)

        for i, ID in enumerate(list_labels_temp):
            temp = self.getImage(ID)
            Y[i,] = self.make3D(temp)
        return X, Y

给定的输出，以等待数小时结束：

Total params: 1,940,817
Trainable params: 1,940,817
Non-trainable params: 0
initialization of DataGen
on_epoch_end()
Epoch 1/30

---

我想这个过程已经陷入僵局了。我在use\u multiprocessing=False下将批处理大小减少到8，并且可以正常工作。但是如果我想使用更大的批量，这仍然不能解决问题。