ValueError: `x` 中的所有数组都应该具有相同的长度。找到一对： len(x[0]) = 97, len(x[?]) = 205

Question

已经检查了this post，但答案没有帮助。

我有以下代码：

LABELS = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}

data = []
labels = []

num_classes = 10
for i in range(num_classes):

    filenames = glob.glob(str(i) + '_*.png')

    for j in range(len(filenames)):
        im_gbr = cv2.imread(filenames[j])
        im = cv2.cvtColor(im_gbr, cv2.COLOR_BGR2RGB)

        data.append(im)
        labels.append(i)

# Normalise
x_min = np.min(data[0], axis=tuple(range(data[0].ndim-1)), keepdims=True)
x_max = np.max(data[0], axis=tuple(range(data[0].ndim-1)), keepdims=True)
data[0] = (data[0] - x_min)/ (x_max - x_min)

lb = LabelBinarizer()
labels = lb.fit_transform(labels)

(trainX, testX, trainY, testY) = train_test_split(data, labels,
                                                  test_size=0.33, stratify=labels, random_state=42)

# construct the training image generator for data augmentation
aug = ImageDataGenerator(rotation_range=20, zoom_range=0.15,
                         width_shift_range=0.2, height_shift_range=0.2, shear_range=0.15,
                         horizontal_flip=True, fill_mode="nearest")

# initialize the optimizer and model
EPOCH = 100
opt = Adam(lr=1e-4, decay=1e-4 / EPOCH)
model = StridedNet.build(width=96, height=96, depth=3,
                         classes=len(lb.classes_), reg=l2(0.0005))
model.compile(loss="categorical_crossentropy", optimizer=opt,
              metrics=["accuracy"])
# train the network

H = model.fit(x=aug.flow(trainX, trainY, batch_size=32),
              validation_data=(testX, testY), steps_per_epoch=len(trainX) // 32,
              epochs=EPOCH)

还有：

len(trainX) = 66

len(testX) = 33

len(trainY) = 66

len(testY) = 33

当我运行代码时，我收到以下错误：

ValueError: All of the arrays in `x` should have the same length. Found a pair with: len(x[0]) = 97, len(x[?]) = 205

错误对应x=aug.flow(trainX, trainY, batch_size=32)

Answer 1

我相信你需要让你所有的图片都一样大小，所以之后

im = cv2.cvtColor(im_gbr, cv2.COLOR_BGR2RGB)
# add the code
height=96     # set your desired image height in pixels
width= 96      # set your desired image width in pixels
im = cv2.resize(im, (width, height))

不确定您为什么要进行标准化。您可以在图像数据生成器中添加 rescale=1/255。当你丢失时，你正在使用 categorical_cross 熵。我认为如果是这种情况，您应该使用 lb=OneHotEncoder(sparse=False)。