我试图在keras中创建一个自定义图层以用于我的VGG模型，但是在打印模型后出现问题。它运作良好并显示模型。但是，当我开始以大于1的批次大小进行训练时，会出现一个错误，如该主题中所示。您能在这方面请任何人帮助我吗？预先感谢。
# include packages
import random
import keras
from numpy.random import seed
seed(1)
random.seed(1)
from keras import backend as K
from keras.models import Model, Sequential
from keras.layers import Flatten, Dense, Dropout, Permute, Lambda, Conv2D, multiply, Concatenate, concatenate, \
RepeatVector
from keras.preprocessing.image import ImageDataGenerator
import numpy as np
import os
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
from keras import optimizers
data_list = os.listdir('D:/xys/X_Ray_Image_DataSet/splits/f2/train')
print(len(data_list))
# Custom layer in our experiment
class Neighbours(keras.layers.Layer):
def __init__(self):
super(Neighbours, self).__init__()
def call(self, vector_):
row_max, col_max = 9, 9
vector = vector_[0]
total_stack_array = []
row_tensor = None
for row_index in range(9):
hit_count = 0
aggregate = K.variable(tf.zeros(shape=(8, 512), dtype=tf.float32))
zeros = K.variable(tf.zeros(shape=(1, 512), dtype=tf.float32))
stack_array = []
for col_index in range(9):
if (row_index - 1) >= 0:
# print("first ", row_index - 1, col_index)
# aggregate[0] = vector[row_index - 1][col_index]
aggregate[0].assign(vector[row_index - 1][col_index])
hit_count += 1
else:
# aggregate[0] = K.zeros(shape=(1, 512))
aggregate[0].assign(zeros)
if (col_index - 1) >= 0:
# print("second ", row_index, col_index - 1)
# aggregate[1] = vector[row_index][col_index - 1]
aggregate[1].assign(vector[row_index][col_index] - 1)
hit_count += 1
else:
# aggregate[1] = K.zeros(shape=(1, 512))
aggregate[1].assign(zeros)
if (row_index - 1) >= 0 and (col_index - 1) >= 0:
# print("third ", row_index - 1, col_index - 1)
# aggregate[2] = vector[row_index - 1][col_index - 1]
aggregate[2].assign(vector[row_index - 1][col_index - 1])
hit_count += 1
else:
# aggregate[2] = K.zeros(shape=(1, 512))
aggregate[2].assign(zeros)
if (row_index + 1) < row_max and (col_index - 1) >= 0:
# print("fourth ", row_index + 1, col_index - 1)
# aggregate[3] = vector[row_index + 1][col_index - 1]
aggregate[3].assign(vector[row_index + 1][col_index - 1])
hit_count += 1
else:
# aggregate[3] = K.zeros(shape=(1, 512))
aggregate[3].assign(zeros)
if (row_index + 1) < row_max:
# print("fifth ", row_index + 1, col_index)
# aggregate[4] = vector[row_index + 1][col_index]
aggregate[4].assign(vector[row_index + 1][col_index])
hit_count += 1
else:
# aggregate[4] = K.zeros(shape=(1, 512))
aggregate[4].assign(zeros)
if (col_index + 1) < col_max:
# print("Sixth ", row_index, col_index + 1)
# aggregate[5] = vector[row_index][col_index + 1]
aggregate[5].assign(vector[row_index][col_index + 1])
hit_count += 1
else:
# aggregate[5] = K.zeros(shape=(1, 512))
aggregate[5].assign(zeros)
hit_count += 1
if (row_index - 1) >= 0 and (col_index + 1) < col_max:
# print("Seventh ", row_index - 1, col_index + 1)
# aggregate[6] = vector[row_index - 1][col_index + 1]
aggregate[6].assign(vector[row_index - 1][col_index + 1])
hit_count += 1
else:
# aggregate[6] = K.zeros(shape=(1, 512))
aggregate[6].assign(zeros)
if (row_index + 1) < row_max and (col_index + 1) < col_max:
# print("Eighth ", row_index + 1, col_index + 1)
# aggregate[7] = vector[row_index + 1][col_index + 1]
aggregate[7].assign(vector[row_index + 1][col_index + 1])
hit_count += 1
else:
# aggregate[7] = K.zeros(shape=(1, 512))
aggregate[7].assign(zeros)
avg = K.sum(aggregate, axis=0) / (hit_count + 1)
stack_array.append(avg)
total_stack_array.append(K.stack(stack_array))
modified_vector = K.stack([K.stack(total_stack_array)])
# print(modified_vector)
aux_array = tf.placeholder_with_default(modified_vector, tf.TensorShape([None, 9, 9, 512]))
return aux_array
root_path = "D:/xys/X_Ray_Image_DataSet/splits/f2/"
DATASET_PATH = root_path + 'train'
test_dir = root_path + 'val'
IMAGE_SIZE = (150, 150)
NUM_CLASSES = len(data_list)
BATCH_SIZE = 1 # try reducing batch size or freeze more layers if your GPU runs out of memory
NUM_EPOCHS = 1
LEARNING_RATE = 0.0001
# Train datagen here is a preprocessor
train_datagen = ImageDataGenerator(rescale=1. / 255,
rotation_range=50,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.25,
zoom_range=0.1,
channel_shift_range=20,
horizontal_flip=True,
vertical_flip=True,
validation_split=0.2,
fill_mode='constant')
# For multiclass use categorical n for binary us
train_batches = train_datagen.flow_from_directory(DATASET_PATH,
target_size=IMAGE_SIZE,
shuffle=True,
batch_size=BATCH_SIZE,
subset="training",
seed=42,
class_mode="categorical"
# For multiclass use categorical n for binary use binary
)
valid_batches = train_datagen.flow_from_directory(DATASET_PATH,
target_size=IMAGE_SIZE,
shuffle=True,
batch_size=BATCH_SIZE,
subset="validation",
seed=42,
class_mode="categorical"
# For multiclass use categorical n for binary use binary
)
from keras.applications import VGG16, VGG19
conv_base = VGG16(weights='imagenet',
include_top=False,
input_shape=(150, 150, 3))
# conv_base.trainable = True
# layer_base=conv_base.output
conv_base = Model(inputs=conv_base.inputs, outputs=conv_base.get_layer('block4_pool').output)
def Trunc_VGG(conv_base):
conv_base.trainable = True
layer_4th = conv_base.get_layer('block4_pool').output
neighbour_layer = Neighbours()
neighbor_module = neighbour_layer(layer_4th)
print(neighbor_module)
flatten = Flatten()(neighbor_module)
dropout = Dropout(0.5)(flatten)
dense = Dense(256, activation='relu')(dropout)
pred = (Dense(3, activation='softmax'))(dense)
model = Model(inputs=conv_base.inputs, outputs=pred)
return model
model = Trunc_VGG(conv_base) # load our model
model.compile(loss='categorical_crossentropy', # for multiclass use categorical_crossentropy
optimizer=optimizers.Adam(lr=LEARNING_RATE),
metrics=['acc'])
print(model.summary())
#
# FIT MODEL
print(len(train_batches))
print(len(valid_batches))
STEP_SIZE_TRAIN = train_batches.n // train_batches.batch_size
STEP_SIZE_VALID = valid_batches.n // valid_batches.batch_size
result = model.fit_generator(train_batches,
steps_per_epoch=STEP_SIZE_TRAIN,
validation_data=valid_batches,
validation_steps=STEP_SIZE_VALID,
epochs=NUM_EPOCHS,
)
import matplotlib
matplotlib.use('TKAgg')
import matplotlib.pyplot as plt
print(result.history.keys())
fig1 = plt.figure(1)
# summarize history for accuracy
plt.plot(result.history['acc'])
plt.plot(result.history['val_acc'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Training', 'Validation'], loc='upper left')
plt.show()
# plt.savefig(root_path + '/' + 'acc.png')
# summarize history for loss
fig2 = plt.figure(2)
plt.plot(result.history['loss'])
plt.plot(result.history['val_loss'])
plt.title('Model Loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Training', 'Validation'], loc='upper left')
plt.show()
# plt.savefig(root_path + '/' + 'loss.png')
test_datagen = ImageDataGenerator(rescale=1. / 255)
eval_generator = test_datagen.flow_from_directory(test_dir, target_size=IMAGE_SIZE, batch_size=1,
shuffle=False, seed=42, class_mode="categorical")
eval_generator.reset()
eval_generator.reset()
x = model.evaluate_generator(eval_generator,
steps=np.ceil(len(eval_generator)),
use_multiprocessing=False,
verbose=1,
workers=1,
)
print('Test loss:', x[0])
print('Test accuracy:', x[1])