ValueError: `.fit()` 的输入应该有 4 级。在 CNN 中有形状的数组吗？

Question

我致力于在我的数据集中实现 CNN。

这是我的代码，通过重塑过程获得 x 训练和 y 训练

Y_train = train["Label"]
X_train = train.drop(labels = ["Label"],axis = 1) 
X_train.shape -> /*(230, 67500)*/
X_train = np.pad(X_train, ((0,0), (0, (67600-X_train.shape[1]))), 'constant').reshape(-1, 260, 260)
Y_train = to_categorical(Y_train, num_classes = 10)

在我完成一些程序和重塑过程之后，我将 X_train 和 Y_train 分开。这是下面显示的代码。

X_train, X_val, Y_train, Y_val = train_test_split(X_train, Y_train, test_size = 0.1, random_state=42)
print("x_train shape",X_train.shape)
print("x_test shape",X_val.shape)
print("y_train shape",Y_train.shape)
print("y_test shape",Y_val.shape)

结果定义如下。

x_train shape (207, 260, 260)
x_test shape (23, 260, 260)
y_train shape (207, 10)
y_test shape (23, 10)

然后我创建 CNN 模型。

model = Sequential()

#
model.add(Conv2D(filters = 8, kernel_size = (5,5),padding = 'Same', 
                 activation ='relu', input_shape = (260, 260)))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.25))

#
model.add(Conv2D(filters = 16, kernel_size = (3,3),padding = 'Same', 
                 activation ='relu'))
model.add(MaxPool2D(pool_size=(2,2), strides=(2,2)))
model.add(Dropout(0.25))

# fully connected
model.add(Flatten())
model.add(Dense(256, activation = "relu"))
model.add(Dropout(0.5))
model.add(Dense(10, activation = "softmax"))

然后我使用 ImageGenerator 来使用数据增强

datagen = ImageDataGenerator(
        featurewise_center=False,  # set input mean to 0 over the dataset
        samplewise_center=False,  # set each sample mean to 0
        featurewise_std_normalization=False,  # divide inputs by std of the dataset
        samplewise_std_normalization=False,  # divide each input by its std
        zca_whitening=False,  # dimesion reduction
        rotation_range=0.5,  # randomly rotate images in the range 5 degrees
        zoom_range = 0.5, # Randomly zoom image 5%
        width_shift_range=0.5,  # randomly shift images horizontally 5%
        height_shift_range=0.5,  # randomly shift images vertically 5%
        horizontal_flip=False,  # randomly flip images
        vertical_flip=False)  # randomly flip images

X_train = np.pad(X_train, ((0,0), (0, (67600-X_train.shape[1]))), 'constant').reshape(-1, 260, 260, 1)

datagen.fit(X_train)

然后它会抛出如下所示的错误。

ValueError: operands could not be broadcast together with remapped shapes [original->remapped]: (2,2) and requested shape (3,2)

我该如何解决？

Answer 1

我认为问题在于ImageDataGenerator 期望图像具有宽度、高度和颜色通道（最常见的是红色、绿色和蓝色的 3 个通道）。由于还有一个批量大小，它期望的整体形状是(batch size, width, height, channels)。您的张量是 260x260，但没有颜色通道。它们是灰度图像吗？

每the documentation:

x：样本数据。应该有等级 4。在灰度数据的情况下， 通道轴的值应为 1

所以我认为你只需要重塑你的输入，在最后添加一个额外的维度。