使用 keras .h5 模型进行预测时出错？

Question

我已经使用图像数据集训练了一个 CNN 模型，并使用名称 classifier.h5 保存了模型。现在，我需要加载这个模型来进行预测。我通过以下方式实现了代码，但是遇到了错误_maybe_load_initial_epoch_from_ckpt() takes 2 positional arguments but 3 were given。什么可能导致此错误以及如何解决？

这是我的 CNN 模型。我将它保存为 classifier.h5

EPOCHS = 60
batch_size = 32
iter_per_epoch = len(x_train)//batch_size
val_per_epoch = len(x_test)//batch_size
print(len(x_train))
print(len(x_test))


classifier = Sequential()
classifier.add(Conv2D(32, 3, activation='relu',
                      padding='same', input_shape=(img_w, img_h, 3)))
classifier.add(BatchNormalization())
classifier.add(MaxPooling2D())

classifier.add(Conv2D(32, 3, activation='relu',
                      padding='same', kernel_initializer='he_uniform'))
classifier.add(BatchNormalization())
classifier.add(MaxPooling2D())

classifier.add(Conv2D(64, 3, activation='relu',
                      padding='same', kernel_initializer='he_uniform'))
classifier.add(BatchNormalization())
classifier.add(MaxPooling2D())

classifier.add(Conv2D(64, 3, activation='relu',
                      padding='same', kernel_initializer='he_uniform'))
classifier.add(BatchNormalization())
classifier.add(MaxPooling2D())

classifier.add(Conv2D(128, 3, activation='relu',
                      padding='same', kernel_initializer='he_uniform'))
classifier.add(BatchNormalization())
classifier.add(MaxPooling2D())

classifier.add(Conv2D(128, 3, activation='relu',
                      padding='same', kernel_initializer='he_uniform'))
classifier.add(BatchNormalization())
classifier.add(MaxPooling2D())

classifier.add(Flatten())
classifier.add(Dropout(0.5))
classifier.add(Dense(128, activation='relu'))
classifier.add(Dense(4, activation='softmax'))

classifier.compile(loss='sparse_categorical_crossentropy',
                   optimizer='adam', metrics=['accuracy'])
classifier.summary()

train_datagen = ImageDataGenerator(
    rotation_range=25,
    zoom_range=0.1,
    width_shift_range=0.1,
    height_shift_range=0.1,
    shear_range=0.2,
    horizontal_flip=True
)

val_datagen = ImageDataGenerator()

train_gen = train_datagen.flow(x_train, y_train, batch_size=batch_size)
val_gen = val_datagen.flow(x_test, y_test, batch_size=batch_size)

m = classifier.fit_generator(
    train_gen,
    steps_per_epoch=iter_per_epoch,
    epochs=EPOCHS,
    validation_data=val_gen,
    validation_steps=val_per_epoch,
    verbose=1
)

现在在另一个脚本中，我正在尝试加载这个模型，但遇到了上述错误

from keras.models import load_model
from tensorflow import Graph
import tensorflow as tf

img_w, img_h = 256, 256
gpuoptions = tf.compat.v1.GPUOptions(allow_growth=True)
graph = Graph()
with graph.as_default():
    tf_session = tf.compat.v1.Session(
        config=tf.compat.v1.ConfigProto(gpu_options=gpuoptions))
    with tf_session.as_default():
        model = load_model('classifier.h5')
path = "./images/test/Leaf Blast/blast__0_2632.jpg"
img = cv2.imread(path)
test_image = cv2.resize(img, (int(img_w*1.5), int(img_h*1.5)))
test_image = preprocess(test_image)
test_image = edge_and_cut(test_image)
test_image = np.array(test_image)
test_image = np.expand_dims(test_image, axis=0)
img_class = model.predict(test_image)
print(img_class)

以下是我的完整错误跟踪：

Traceback (most recent call last):
  File "error.py", line 171, in <module>
    img_class = model.predict(test_image)
  File "C:\Users\namba\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\keras\engine\training_v1.py", line 983, in predict
    return func.predict(
  File "C:\Users\namba\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\keras\engine\training_arrays.py", line 708, in predict
    return predict_loop(
  File "C:\Users\namba\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\keras\engine\training_arrays.py", line 259, in model_iteration
    initial_epoch = model._maybe_load_initial_epoch_from_ckpt(initial_epoch, mode)
TypeError: _maybe_load_initial_epoch_from_ckpt() takes 2 positional arguments but 3 were given

Answer 1

我不知道是什么导致了错误。但是，以下更改已解决该问题。我删除了 Graph 部分并加载了我的模型。现在它正在显示结果。

class_dict = {'Bacterial leaf blight': 0,
              'Brown spot': 1,
              'Leaf Blast': 2,
              'Leaf smut': 3
              }
class_names = list(class_dict.keys())

img_w, img_h = 256, 256
model = load_model('classifier.h5')
path = "./images/test/Leaf Blast/blast__0_2632.jpg"
img = cv2.imread(path)
test_image = cv2.resize(img, (int(img_w*1.5), int(img_h*1.5)))
test_image = preprocess(test_image)
test_image = edge_and_cut(test_image)
test_image = np.array(test_image)
test_image = np.expand_dims(test_image, axis=0)
img_class = model.predict(test_image)
img_class = img_class.flatten()
m = max(img_class)
for index, item in enumerate(img_class):
    if item == m:
        pred_class = class_names[index]
print(pred_class)