Question

我有两类疾病A和B。我的训练数据包含28张图片，包括两个类别。我已经使用opencv创建了大小调整功能。

def resize_cv(x,width,height):
    new_image=cv.resize(x,(width,height))

    return new_image

X包含28张图像的列表。

xx=[]
for i in X:
    xx.append(resize_cv(i,196,196))                                                 #resizing happens here
print("__Resized the images__")

def scaling (X):
    new=[]
    for i in X:
        for j in i:
            new.append(j/255)
            break
    return new

def label_encode(y):
    from sklearn.preprocessing import LabelBinarizer
    ff=LabelBinarizer()
    return ff.fit_transform(y)


X=scaling(xx)
y=label_encode(y)

现在我将数据分为训练和测试并创建步长

X_train,X_test, y_train, y_test=split_data(X,y,0.2)
#creating smaller batches 
step_size=7
steps = len(X_train)
remaining = steps % step_size

现在，当我进入神经网络时，我已经创建了一个神经网络。

layer_conv1 = create_convolutional_layer(input=x,num_input_channels=num_channels,conv_filter_size=filter_size_conv1,num_filters=num_filters_conv1,name="conv1")

layer_conv1_1 = create_convolutional_layer(input=layer_conv1,num_input_channels=num_filters_conv1,conv_filter_size=filter_size_conv1,num_filters=num_filters_conv1,name="conv2")

layer_conv1_1_1 = create_convolutional_layer(input=layer_conv1_1,num_input_channels=num_filters_conv1,conv_filter_size=filter_size_conv1,num_filters=num_filters_conv1,name="conv3")

max_pool_1=maxpool2d(layer_conv1_1_1,2,name="maxpool_1")

drop_out_1=dropout(max_pool_1,name="dropout_1")

flatten_layer=create_flatten_layer(drop_out_3)

layer_fc2 = create_fc_layer(input=flatten_layer,num_inputs=fc_layer_size,num_outputs=num_classes,use_relu=True)


y_pred = tf.nn.softmax(layer_fc2,name="y_pred")

cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y,logits=y_pred))

#Defining objective
train = tf.train.AdamOptimizer(learning_rate=0.00001).minimize(cost)


print ("_____Neural Network Architecture Created Succefully_____")
epochs=10
    matches = tf.equal(tf.argmax(y_pred,axis=1),tf.argmax(y,axis=1))
acc = tf.reduce_mean(tf.cast(matches,tf.float32))


#Initializing weights
init = tf.global_variables_initializer()

with tf.Session() as sess:
    #writing output to the logs for tensorboard 
    writer=tf.summary.FileWriter("./logs",sess.graph)
    sess.run(init)

    for i in range(epochs):
        #creating smaller batches

        for j in  range(0,steps-remaining,step_size):
            sess.run([acc,train,cost],feed_dict={x:X_train[j:j+step_size],y:y_train[j:j+step_size]})

错误跟踪：

Traceback (most recent call last):
  File "/home/centura/gitlab/moles_model/moles_model/modelversion1.py", line 313, in <module>
    sess.run([acc,train,cost],feed_dict={x:X_train,y:y_train})
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 900, in run
    run_metadata_ptr)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1111, in _run
    str(subfeed_t.get_shape())))
ValueError: Cannot feed value of shape (22, 196, 3) for Tensor 'x:0', which has shape '(?, 196, 196, 3)'

我已经检查了 X 数组中的图片尺寸。每个图像都保持尺寸（196,196,3），但是

当我在 X_train 中检查图像的尺寸时，每个图像的尺寸为（196,3）。

我没到失踪的地方 196 去。

我正在使用tensorflow-gpu=1.9.0, python 3.6 , pycharm IDE.

Answer 1

答案很简单，原因是（196,196,3）的转换归因于缩放功能中的额外for循环。

代替使用此代码

def scaling (X):
    new=[]
    for i in X:
        for j in i:
            new.append(j/255)
            break
    return new

我应该避免第二个循环，函数看起来像这样：

def scaling (X):
    new=[]
    for i in X:
        new.append(i/255)
    return new

Tensorflow feed_dict维度未与神经网络输入和训练输入匹配

1 个答案: