tensorflow特征图可视化

代码是基于《实战GOOGLE深度学习框架》书中手写数字识别的示例，
做了一些修改

inference函数最后的返回处加上某一层，比如conv1

修改train函数中y=inference()

                a,b,c,d,e=sess.run([conv1_img,pool1_img,conv2_img,conv3_img,pool3_img],feed_dict={x:_x_})     
#                a=sess.run(conv1_img,feed_dict={x:_x_})
#                print(type(a))
#                print(a.shape)
#                print(a[0,:,:,0].shape)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv1_0\{}.jpg'.format(i),a[0,:,:,0]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv1_1\{}.jpg'.format(i),a[0,:,:,1]*255)

inference函数返回的特征图是数据格式是tensor，不能直接保存，所以用sess.run()来将数据格式转换为ndarray来保存

def inference(input_tensor,train,regularizer):
    with tf.variable_scope('layer1-conv1'):
        conv1_weights=tf.get_variable('weights',[CONV1_SIZE,CONV1_SIZE,NUM_CHANNELS,CONV1_DEEP],initializer=tf.truncated_normal_initializer(stddev=0.1))
        conv1_biases=tf.get_variable('biases',[CONV1_DEEP],initializer=tf.constant_initializer(0.0))
        
        conv1=tf.nn.conv2d(input_tensor,conv1_weights,strides=[1,1,1,1],padding='SAME')
        relu1=tf.nn.relu(tf.nn.bias_add(conv1,conv1_biases))
        
    with tf.name_scope('layer2-pool1'):
        pool1=tf.nn.max_pool(relu1,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')
        
#        print(pool1)
        
    with tf.variable_scope('layer3-conv2'):
        conv2_weights=tf.get_variable('weights',[CONV2_SIZE,CONV2_SIZE,CONV1_DEEP,CONV2_DEEP],initializer=tf.truncated_normal_initializer(stddev=0.1))
        conv2_biases=tf.get_variable('biases',[CONV2_DEEP],initializer=tf.constant_initializer(0.0))
        
        conv2=tf.nn.conv2d(pool1,conv2_weights,strides=[1,1,1,1],padding='SAME')
        relu2=tf.nn.relu(tf.nn.bias_add(conv2,conv2_biases))
        
#    with tf.name_scope('layer4-pool2'):
#        pool2=tf.nn.max_pool(relu2,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')


    with tf.variable_scope('layer5-conv2'):
        conv3_weights=tf.get_variable('weights',[CONV3_SIZE,CONV3_SIZE,CONV2_DEEP,CONV3_DEEP],initializer=tf.truncated_normal_initializer(stddev=0.1))
        conv3_biases=tf.get_variable('biases',[CONV3_DEEP],initializer=tf.constant_initializer(0.0))
        
        conv3=tf.nn.conv2d(relu2,conv3_weights,strides=[1,1,1,1],padding='SAME')
        relu3=tf.nn.relu(tf.nn.bias_add(conv3,conv3_biases))
        
    with tf.name_scope('layer6-pool2'):
        pool3=tf.nn.max_pool(relu3,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')        
        
        
        
    pool_shape=pool3.get_shape().as_list()
    nodes=pool_shape[1]*pool_shape[2]*pool_shape[3]
    
    reshaped=tf.reshape(pool3,[pool_shape[0],nodes])
    
    with tf.variable_scope('layer5-fc1'):
        fc1_weights=tf.get_variable('weights',[nodes,FC_SIZE],initializer=tf.truncated_normal_initializer(stddev=0.1))
        
        if regularizer !=None:
            tf.add_to_collection('loses',regularizer(fc1_weights))
        fc1_biases=tf.get_variable('biases',[FC_SIZE],initializer=tf.constant_initializer(0.1))
        
        fc1=tf.nn.relu(tf.matmul(reshaped,fc1_weights)+fc1_biases)
 # 如果train标签为真，则引入dropout函数使输出层一半的神经元失活
        if train:fc1=tf.nn.dropout(fc1,0.5)
        
    with tf.variable_scope('layer6-fc2'):
        fc2_weights=tf.get_variable('weights',[FC_SIZE,NUM_LABELS],initializer=tf.truncated_normal_initializer(stddev=0.1))
        if regularizer !=None:
            tf.add_to_collection('losses',regularizer(fc2_weights))
            
        fc2_biases=tf.get_variable('biases',[NUM_LABELS],initializer=tf.constant_initializer(0.1))
        logit=tf.matmul(fc1,fc2_weights)+fc2_biases
      
        
    return logit,conv1,pool1,conv2,conv3,pool3
    
#    
#def get_weight_variable(shape,regularizer):
#    weights=tf.get_variable('weights',shape,initializer=tf.truncated_normal_initializer(stddev=0.1))
#    if regularizer != None:
#        tf.add_to_collection('losses',regularizer(weights))

#def conv_img():
    



#定义训练过程
def train(mnist):
#    输入数据,输出数据
    x=tf.placeholder(tf.float32,[BATCH_SIZE,INPUT_NODE],name='x-input')
    reshaped_xs=tf.reshape(x, (BATCH_SIZE, 28, 28,1))
    y_=tf.placeholder(tf.float32,[None,OUTPUT_NODE],name='y-input')
    
    regularizer=tf.contrib.layers.l2_regularizer(REGULARAZTION_RATE)
# 计算神经网络在当前参数下的前向传播结果   
    y,conv1_img,pool1_img,conv2_img,conv3_img,pool3_img=inference(reshaped_xs,False,regularizer)
#    训练轮数变量
    global_step=tf.Variable(0,trainable=False)
#    初始化滑动平均
    variable_averages=tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY,global_step)
#tf.get_variables返回所有没有指定trainable=False的参数
#    在所有代表神经网络参数的变量上使用滑动平均值
    variables_averages_op=variable_averages.apply(tf.trainable_variables())
#  计算交差熵  ，tf.argmax获得正确答案对应编号
    cross_entropy=tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y,labels=tf.argmax(y_,1))
#   计算在当前batch中所有样例交差熵的平均值 
    cross_entropy_mean=tf.reduce_mean(cross_entropy)
    
    loss=cross_entropy_mean+tf.add_n(tf.get_collection('losses'))
#    指数衰减学习率
    learning_rate=tf.train.exponential_decay(LEARNING_RATE_BASE,global_step,100,LEARNING_RATE_DECAY)
#    优化损失函数
    train_step=tf.train.GradientDescentOptimizer(learning_rate)\
                 .minimize(loss,global_step=global_step)
#    反向传播，更新神经网络中参数和每一个参数的滑动平均值
                 
    with tf.control_dependencies([train_step,variables_averages_op]):
        train_op=tf.no_op(name='train')
        
    saver=tf.train.Saver()
    
    correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
    accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
    
    with tf.Session() as sess:
        tf.global_variables_initializer().run()
#      测试数据  
#        test_feed={x:mnist.test.next_batch(BATCH_SIZE),y_:mnist.test.next_batch(BATCH_SIZE)}
#   迭代地训练神经网络
        start=time.clock()
        for i in range(TRAINING_STEPS):
#            产生这一轮使用的一个batch的训练数据，并运行训练
            xs,ys=mnist.train.next_batch(BATCH_SIZE)
            
            _,loss_value,step=sess.run([train_op,loss,global_step],feed_dict={x:xs,y_:ys})
            
            if i % 10 ==0:
                
                _x_,_y_=mnist.validation.next_batch(BATCH_SIZE)
                
#                print(_x_.shape)
                
#                reshaped_x_=np.reshape(_x_, (BATCH_SIZE, 28, 28,1))

#                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\2.jpg',reshaped_x_[0,:,:]*255)



#                print(conv1_img)

                a,b,c,d,e=sess.run([conv1_img,pool1_img,conv2_img,conv3_img,pool3_img],feed_dict={x:_x_})     
#                a=sess.run(conv1_img,feed_dict={x:_x_})
#                print(type(a))
#                print(a.shape)
#                print(a[0,:,:,0].shape)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv1_0\{}.jpg'.format(i),a[0,:,:,0]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv1_1\{}.jpg'.format(i),a[0,:,:,1]*255)
                
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv2_0\{}.jpg'.format(i),c[0,:,:,0]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv2_1\{}.jpg'.format(i),c[0,:,:,1]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv2_2\{}.jpg'.format(i),c[0,:,:,2]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv2_3\{}.jpg'.format(i),c[0,:,:,3]*255)    
    
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv3_0\{}.jpg'.format(i),d[0,:,:,0]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv3_1\{}.jpg'.format(i),d[0,:,:,1]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv3_2\{}.jpg'.format(i),d[0,:,:,2]*255)
                cv2.imwrite(r'C:\Users\Administrator\Desktop\test\conv3_3\{}.jpg'.format(i),d[0,:,:,3]*255)                
#                sns.heatmap(a[0,:,:,1])   #热力图
                
                validate_acc=sess.run(accuracy,feed_dict={x:_x_,y_:_y_})
            
                print('after %d training steps, validation accuracy ' 'using average model is %g ' % (step,validate_acc))
                
                print ('after %d training steps, loss on training ' 'batch is %g.' % (step,loss_value))
                
                with open(r'C:\Users\Administrator\Desktop\tensorflow\mnist17.txt','a') as f:
                    f.writelines([str(step),'#','#',str(loss_value),'#',str(validate_acc),'\n'])
#                str(validate_acc),
#                DATA_X.append(i/5)
#                DATA_Y.append(loss_value)
                
                saver.save(sess,os.path.join(MODEL_SAVE_PATH,MODEL_NAME),global_step=global_step)
        
#                test_acc=sess.run(accuracy,feed_dict=test_feed)
#                print('after %d training steps, test accuracy using average' 'model is %g ' % (TRAINING_STEPS,test_acc))
                
        end=time.clock() 
        print("running time is %g s" % (end-start))