Question

我在这个主题上经历了所有的问题。这里遇到一个奇怪的问题。我将图像路径存储在file_names。

中

from skimage import io
import numpy as np

X = np.array([np.array(io.imread(i)) for i in file_names])
print(X.shape)
# (50,)
print(X[0].shape)
# (375, 500, 3)

我需要X为(50, 375, 500, 3)。我尝试reshape，添加np.newaxis等但都失败了。我的下一步是将其用于CNN。基本上，我想用我的图像创建一个mnist_cnn类数据集。

下一行：

model = Sequential()
model.add(Conv2D(64, kernel_size=(3, 3),
                 activation='relu',
                 input_shape = (375, 500, 3)))
model.add(Flatten())
model.add(Dense(num_classes, activation='softmax'))

model.compile(loss='categorical_crossentropy',
              optimizer='adam', metrics=['accuracy'])
model.fit(X, y,   # y is (50,36) using one hot encoding
          batch_size=10,
          epochs=10,
          verbose=2)

原因：

ValueError: Error when checking input: expected conv2d_3_input to have 4 dimensions, but got array with shape (50, 1)

Answer 1

numpy-part看起来很简单：

from skimage import io
import numpy as np

# assumption: images are homogeneous in terms of dimensions and channels!
files = ['C:/TEMP/pic0.jpg', 'C:/TEMP/pic0.jpg', 'C:/TEMP/pic0.jpg', 'C:/TEMP/pic0.jpg']

image_array = np.stack([io.imread(i) for i in files])                  # default: axis=0
image_array.shape
# (4, 720, 540, 3)

无法将RGB图像读入numpy数组

1 个答案: