使用Keras进行图像分类："expected activation_1 to have shape (2,) but got array with (1,)"

Question

我正在训练一个图像分类器来区分猫和狗，来自Kaggle集。

这是我的相关代码：

FINAL_ACTIVATION = "softmax"
OPTIMIZER = keras.optimizers.Adamax()
STRIDES = (2, 2)
DROPOUT = 0.5
model = Sequential()
model.add(Conv2D(32, (3, 3), input_shape = INPUT_SHAPE))
model.add(keras.layers.ELU())
model.add(MaxPooling2D(pool_size = (2, 2), strides = STRIDES))
model.add(Conv2D(64, (3, 3)))
model.add(keras.layers.ELU())
model.add(MaxPooling2D(pool_size = (2, 2), strides = STRIDES))
model.add(Conv2D(128, (3, 3)))
model.add(keras.layers.ELU())
model.add(MaxPooling2D(pool_size = (2, 2), strides = STRIDES))
model.add(Conv2D(128, (3, 3)))
model.add(keras.layers.ELU())
model.add(MaxPooling2D(pool_size = (2, 2), strides = STRIDES))
model.add(Flatten())
model.add(Dense(100))
model.add(keras.layers.ELU())
model.add(Dropout(DROPOUT))
model.add(Dense(50))
model.add(keras.layers.ELU())
model.add(Dropout(DROPOUT))
model.add(Dense(2))
model.add(Activation(FINAL_ACTIVATION))
model.compile(
    loss="categorical_crossentropy",
    optimizer = OPTIMIZER,
    metrics = ["accuracy"]
)
train_datagen = ImageDataGenerator(
    rescale = 1./255,
    shear_range = 0.2,
    zoom_range = 0.2,
    horizontal_flip = True
)
train_generator = train_datagen.flow(
    x_train,
    y_train,
    batch_size = BATCH_SIZE
)
model.summary()
model.fit_generator(
    train_generator,
    steps_per_epoch = x_train.shape[0] // BATCH_SIZE,
    epochs = EPOCHS
)

由于训练文件数量众多，因此没有测试集。但是，当我尝试训练这个网络时，我收到错误"ValueError：检查目标时出错：预期activation_1具有形状 (2，(，但得到具有形状 (1，( 的数组。我在这里做错了什么？

卷积神经网络对我来说仍然有点黑魔法，所以我可能在这里犯了一些初学者的错误。我想我可能是。

这是我的模型：

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
conv2d_1 (Conv2D)            (None, 98, 98, 32)        896
_________________________________________________________________
elu_1 (ELU)                  (None, 98, 98, 32)        0
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 49, 49, 32)        0
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 47, 47, 64)        18496
_________________________________________________________________
elu_2 (ELU)                  (None, 47, 47, 64)        0
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 23, 23, 64)        0
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 21, 21, 128)       73856
_________________________________________________________________
elu_3 (ELU)                  (None, 21, 21, 128)       0
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 10, 10, 128)       0
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 8, 8, 128)         147584
_________________________________________________________________
elu_4 (ELU)                  (None, 8, 8, 128)         0
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 4, 4, 128)         0
_________________________________________________________________
flatten_1 (Flatten)          (None, 2048)              0
_________________________________________________________________
dense_1 (Dense)              (None, 100)               204900
_________________________________________________________________
elu_5 (ELU)                  (None, 100)               0
_________________________________________________________________
dropout_1 (Dropout)          (None, 100)               0
_________________________________________________________________
dense_2 (Dense)              (None, 50)                5050
_________________________________________________________________
elu_6 (ELU)                  (None, 50)                0
_________________________________________________________________
dropout_2 (Dropout)          (None, 50)                0
_________________________________________________________________
dense_3 (Dense)              (None, 2)                 102
_________________________________________________________________
activation_1 (Activation)    (None, 2)                 0
=================================================================
Total params: 450,884
Trainable params: 450,884
Non-trainable params: 0
_________________________________________________________________

Answer 1

由于训练文件数量众多，因此没有测试集。

虽然这不是你的问题，但我觉得应该解决这个问题：我知道有些小组遵循非常不同的方法并且不保留有效/测试集，但这对我来说听起来非常非正统和不确定。就个人而言，我很少看到没有测试测试或某种训练后验证的文章。

测试集非常有用，因为它们是模型适用于看不见的样本的经验证据。使用它们 + 统计显著性/置信度检验，您可以根据统计基础断言您的模型不仅是一个非常复杂的 if-else 机器。

最重要的是，如果您是该地区的新手，则不应遵守此类冒险策略。

关于您的问题：y_train.shape是什么？我相信您正在流动稀疏标签([1., 0., 1., 2., 3., ..., 4.](而不是独热编码的标签：

[[0. 1.]
 [1. 0.]
 [0. 1.]
 ...
 [1. 0.]]

argmax(x, axis=1)为您提供真正的类索引。

如何解决：

将loss="categorical_crossentropy"替换为 sparse_categorical_crossentropy 。这是内存效率高的，但要求每个样本与单个标签相关联。

或

在将y_train传递给ImageDataGenerator#flow之前对其进行单热编码：

from keras.utils import to_categorical
y_train = to_categorical(y_train)
train_datagen = ImageDataGenerator(...)
train_generator = train_datagen.flow(x_train, y_train, ...)
...

这会占用更多内存，但允许单个样本与多个标签相关联：

[[1. 1. 0.]
 [0. 0. 1.]
 ...
 [1. 0. 1.]]