ValueError：检查输入时出错：预期 input_1 有 4 个维度，但得到了形状为 (6243, 256, 256) 的数组

Question

我想在训练数据集上附加标签，我这样做

def one_hot_label(img):
    label = img
    if label == 'A':
        ohl = np.array([1, 0])
    elif label == 'B':
        ohl = np.array([0, 1])
    return ohl

def train_data_with_label():
    train_images = []
    for i in tqdm(os.listdir(train_data)):
        path_pre = os.path.join(train_data, i)
        for img in os.listdir(path_pre):
            if img.endswith('.jpg'):
                path = os.path.join(path_pre, img)
                img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
                train_images.append([np.array(img), one_hot_label(i)])
    shuffle(train_images)
    return train_images

但是，在Keras上执行输入时返回的错误

training_images = train_data_with_label()
tr_img_data = np.array([i[0] for i in training_images])
tr_lbl_data = np.array([i[1] for i in training_images])

model = Sequential()
model.add(InputLayer(input_shape=(256, 256, 1)))

任何人都可以帮我修复它吗？

Answer 1

您的输入层需要一个 shape 数组，(batch_size, 256, 256, 1)但看起来您正在传递 shape 的数据(batch_size, 256, 256)。您可以尝试按如下方式重塑训练数据：

tr_img_data = np.expand_dims(tr_img_data, axis=-1)