Question

这里，我用激活函数relu和softmax在一个隐藏层上训练我的顺序模型。我还仔细检查了数据集图像，然后输入相同类型的图像以通过简单的模型进行预测。但是它无法正确预测。如果准确性很高，那么为什么我得到错误的预测也无法理解。

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

    x_train = x_train / 255.0
    x_train = x_train.reshape(-1,28,28,1)

    model =tf.keras.Sequential([
            tf.keras.layers.Flatten(),
            tf.keras.layers.Dense(512,activation=tf.nn.relu),
            tf.keras.layers.Dropout(0.2),
            tf.keras.layers.Dense(10,activation=tf.nn.softmax)
            ])

    model.compile(optimizer='adam', 
                  loss='sparse_categorical_crossentropy', 
                  metrics=['accuracy'])
    model.fit(x=x_train,y=y_train, epochs=40)

    def input_image(filepath):
        img_size = 28  
        # read in the image and convert to grayscale
        img_array = cv2.imread(filepath, cv2.IMREAD_GRAYSCALE) 

         # resize image to match model's expected size
        new_array = cv2.resize(img_array, (img_size, img_size)) 

        # return the image with shaping that TF wants
        return new_array.reshape(-1, img_size, img_size, 1) 

    #here call the input_image function which return an nested list
    prediction = model.predict([input_image("/home/farhana/Desktop/image processing/code/2.png")]) 

    print(prediction)

   print(np.argmax(prediction))

Answer 1

我认为问题可能在于您没有在input_image函数中规范测试数据。尝试添加一条行，在该行中，将加载的图像除以255.0，就像在训练集中的情况一样。请参见下面的代码：

def input_image(filepath):
    img_size = 28  
    img_array = cv2.imread(filepath, cv2.IMREAD_GRAYSCALE) 
    new_array = cv2.resize(img_array, (img_size, img_size))
    new_array = new_array / 255.0

    return new_array.reshape(-1, img_size, img_size, 1)

我使用TensoFlow 2.0和python3.6以99.68％的精度训练MNIST数据集，但它预测数字错误

1 个答案: