Keras CNN 模型精度保持相对不变，val_accuracy 没有提高

Question

我正在尝试使用 Keras 训练一个模型来识别恶性和良性图像，但是我没有达到我希望的结果。数据集分类良好，从 ISIC - Archive (https://www.isic-archive.com/) 收集。我曾多次尝试更改学习率，但无济于事...results from one of the training intervals

下面是我使用 Adam Optimizer 训练模型的代码：

# In[1]:

from keras.callbacks import ModelCheckpoint
from keras.models import Sequential
from keras.layers import Dropout, Flatten, Dense
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from PIL import ImageFile
from tqdm import tqdm
from keras.preprocessing import image
from sklearn.datasets import load_files
from keras.utils import np_utils
import numpy as np
from glob import glob
import keras

# define function to load train, test, and validation datasets


def load_dataset(path):
    data = load_files(path)

    condition_files = np.array(data['filenames'])

    condition_targets = np_utils.to_categorical(np.array(data['target']), 2)
    print(condition_targets)
    return condition_files, condition_targets


# load train, test, and validation datasets
train_files, train_targets = load_dataset(
    '/Users/Grampun/Desktop/ISIC-Archive-Downloader-master/data_set/training_data')
valid_files, valid_targets = load_dataset(
    '/Users/Grampun/Desktop/ISIC-Archive-Downloader-master/data_set/valid_data')
test_files, test_targets = load_dataset(
    '/Users/Grampun/Desktop/ISIC-Archive-Downloader-master/data_set/test_data')

# load list of labels
condition_names = [item[58:-1] for item in sorted(
    glob("/Users/Grampun/Desktop/ISIC-Archive-Downloader-master/data_set/training_data/*/"))]
print(condition_names)
# print statistics about the dataset
print('There are %d total categories.' % len(condition_names))
print('There are %s total images.\n' %
      len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training images.' % len(train_files))
print('There are %d validation images.' % len(valid_files))
print('There are %d test images.' % len(test_files))


# In[2]:

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)


def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path)
                       for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)


ImageFile.LOAD_TRUNCATED_IMAGES = True

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255


# (IMPLEMENTATION) Model Architecture
#

# In[4]:

model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation='relu',
                 input_shape=(224, 224, 3)))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(2, activation='sigmoid'))

model.summary()


#  Compile the Model

# In[ ]:
# opt = keras.optimizers.Adadelta()
opt = keras.optimizers.Adam(lr=0.00003, beta_1=0.9,
                            beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])


# ### Train the Model
#

# In[ ]:


# TODO: specify the number of epochs that you would like to use to train the model.

epochs = 20

checkpointer = ModelCheckpoint(filepath='weights.best.from_scratch.6.hdf5',
                               verbose=1, save_best_only=True)

model.fit(train_tensors, train_targets,
          validation_data=(valid_tensors, valid_targets),
          epochs=epochs, batch_size=10, callbacks=[checkpointer], verbose=1)

# ### Load the Model with the Best Validation Loss

# In[5]:

model.load_weights('weights.best.from_scratch.6.hdf5')


# ### Test the Model
#

# In[6]:

# get index of predicted label for each image in test set
condition_predictions = [np.argmax(model.predict(
    np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(condition_predictions) ==
                           np.argmax(test_targets, axis=1))/len(condition_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)  # confusion matrix

对此的任何帮助将不胜感激（这是我的第一个 ML 项目，我仍在学习）。 谢谢！

Answer 1

这行代码是你的问题根源：model.add(Dense(2, activation='sigmoid'))。

任意使用：

1. model.add(Dense(2, activation='softmax'))
2. model.add(Dense(1, activation='sigmoid'))

请注意，在情况 (1) 中，您需要使用“categorical_crossentropy”而不是“binary_crossentropy”。因此，你也必须改变

model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])到

model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])