Question

我正在遵循Rajsha编写的指南： https://github.com/rajshah4/image_keras/blob/master/notebook_extras.ipynb

这个想法是将VGG16应用于我的由频谱图组成的数据集，并让其在正常和异常两类之间进行决策。

但是，该模型没有学习，尽管我处于顶层，但我仍然获得了大约0.5 val_acc。

我做错什么了吗？我将代码留在下面：

# dimensions of our images
img_width, img_height = 240, 240

train_data_dir = '/content/gdrive/My Drive/Melspec/melspecimages/train'
validation_data_dir = '/content/gdrive/My Drive/Melspec/melspecimages/val'

batch_size = 32
datagen = ImageDataGenerator(preprocessing_function=preprocess_input)

model_vgg = applications.VGG16(include_top=False, weights='imagenet',input_shape=(240,240,3))

model_vgg.trainable=False

train_generator_bottleneck = datagen.flow_from_directory(
        train_data_dir,
        target_size=(img_width, img_height),
        batch_size=batch_size,
        class_mode='binary',
        shuffle=True)

validation_generator_bottleneck = datagen.flow_from_directory(
        validation_data_dir,
        target_size=(img_width, img_height),
        batch_size=batch_size,
        class_mode='binary',
        shuffle=False) 

train_samples = 30272
validation_samples = 7584

bottleneck_features_train = model_vgg.predict_generator(train_generator_bottleneck, train_samples // batch_size)
np.save(open('/content/gdrive/My Drive/Melspec/spec_vgg_bottleneck_features_train.npy', 'wb'), bottleneck_features_train)

bottleneck_features_validation = model_vgg.predict_generator(validation_generator_bottleneck, validation_samples // batch_size)
np.save(open('/content/gdrive/My Drive/Melspec/spec_vgg_bottleneck_features_validation.npy', 'wb'), bottleneck_features_validation)

train_data = np.load(open('/content/gdrive/My Drive/Melspec/spec_vgg_bottleneck_features_train.npy', 'rb'))
train_labels = np.array([0] * (train_samples // 2) + [1] * (train_samples // 2))

validation_data = np.load(open('/content/gdrive/My Drive/Melspec/spec_vgg_bottleneck_features_validation.npy', 'rb'))
validation_labels = np.array([0] * (validation_samples // 2) + [1] * (validation_samples // 2))

model_top = Sequential()
model_top.add(Flatten(input_shape=train_data.shape[1:]))
model_top.add(Dense(256, activation='relu'))
model_top.add(Dropout(0.5))
model_top.add(Dense(1, activation='sigmoid'))

model_top.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['accuracy'])

model_top.fit(train_data, train_labels,
        epochs=epochs, 
        batch_size=batch_size,
        validation_data=(validation_data, validation_labels))
```

Answer 1

找到答案：我的标签有误。

我在网络上读到，在给train_generator供电时，我们应该使用shuffle = True，但是这些类不是以相同的顺序混合的，只有文件混合在一起，从而导致错误的标签。

我改用shuffle = False，也改用class_mode = None。

我还必须确保数据库中的文件在两个类中的编号相同，并且可以被我的batch_size整除。

希望这对其他初学者有所帮助！

频谱数据集上的VGG16

1 个答案: