Skip to main content
 首页 » 编程设计

python CNN。如何获取数据的隐藏fc层编码python

2025年01月19日10lori

我是 python 的新手,我正在尝试对我的个人图像数据运行 CNN。我得到了合理的结果,但我的代码非常粗糙。

目前我有两个问题: 1) 我想使用最后一个隐藏的 fc 层作为数据的编码,并在"new"数据集上使用其他分类方法,我该怎么做? 2) 如何自动计算第一个 fc 层的输入大小?我有 48x64 图像,当我更改架构中的某些内容时,我必须手动更改 variables_lenet5 方法中的 w3 值。

我正在使用 TensorFlow。这是我脚本的核心:

def variables_lenet5(filter_size = filter_size_1, 
                 filter_size_2=filter_size_2,  
                 filter_depth1 = filter_size_1,  
                 filter_depth2 = filter_size_2, 
                 num_hidden1 = hid_1,  
                 num_hidden2 = hid_2, 
                 #num_hidden3 = hid_3, 
                 image_width = image_width,  
                 image_height = image_height,  
                 image_depth = 3,  
                 num_labels = num_labels): 
 
w1 = tf.Variable(tf.truncated_normal([filter_size, filter_size, image_depth, filter_depth1], stddev=0.1)) 
b1 = tf.Variable(tf.zeros([filter_depth1])) 
 
w2 = tf.Variable(tf.truncated_normal([filter_size_2, filter_size_2, filter_depth1, filter_depth2], stddev=0.1)) 
b2 = tf.Variable(tf.constant(1.0, shape=[filter_depth2])) 
 
#w2_a = tf.Variable(tf.truncated_normal([num_hidden3,num_labels], stddev=0.1)) 
#b2_a = tf.Variable(tf.constant(1.0, shape=[num_labels])) 
 
w3 = tf.Variable(tf.truncated_normal([288, num_hidden1], stddev=0.1))  
b3 = tf.Variable(tf.constant(1.0, shape = [num_hidden1])) 
 
w4 = tf.Variable(tf.truncated_normal([num_hidden1, num_hidden2], stddev=0.1)) 
b4 = tf.Variable(tf.constant(1.0, shape = [num_hidden2])) 
 
w5 = tf.Variable(tf.truncated_normal([num_hidden2, num_labels], stddev=0.1)) 
b5 = tf.Variable(tf.constant(1.0, shape = [num_labels])) 
 
#w5 = tf.Variable(tf.truncated_normal([num_hidden2, num_hidden3], stddev=0.1)) 
#b5 = tf.Variable(tf.constant(1.0, shape = [num_hidden3])) 
variables = { 
    'w1': w1, 'w2': w2, 'w3': w3, 'w4': w4, 'w5': w5, #'w6':w2_a, 
    'b1': b1, 'b2': b2, 'b3': b3, 'b4': b4, 'b5': b5, #'b6':b2_a 
} 
return variables 
 
def model_lenet5(data, variables): 
    layer1_conv = tf.nn.conv2d(data, variables['w1'], [1, 1, 1, 1], padding='SAME') 
    layer1_actv = tf.sigmoid(layer1_conv + variables['b1']) 
    layer1_pool = tf.nn.avg_pool(layer1_actv, [1, 2, 2, 1], [1, 2, 2, 1], padding='SAME') 
 
    layer2_conv = tf.nn.conv2d(layer1_pool, variables['w2'], [1, 2, 2, 1], padding='SAME') #era Valid 
    layer2_actv = tf.sigmoid(layer2_conv + variables['b2']) 
    layer2_pool = tf.nn.max_pool(layer2_actv, [1, 2, 2, 1], [1, 2, 2, 1], padding='SAME') 
 
   # layer3_conv = tf.nn.conv2d(layer2_pool, variables['w6'], [1, 2, 2, 1], padding='VALID') #era Valid 
    #layer3_actv = tf.sigmoid(layer3_conv + variables['b6']) 
    #layer3_pool = tf.nn.max_pool(layer3_actv, [1, 2, 2, 1], [1, 2, 2, 1], padding='SAME') 
 
    flat_layer = flatten_tf_array(layer2_pool) 
    layer3_fccd = tf.matmul(flat_layer, variables['w3']) + variables['b3'] 
    layer3_actv = tf.nn.sigmoid(layer3_fccd) 
 
    layer4_fccd = tf.matmul(layer3_actv, variables['w4']) + variables['b4'] 
    layer4_actv = tf.nn.sigmoid(layer4_fccd) 
    #logits = tf.matmul(layer4_actv, variables['w5']) + variables['b5'] 
    logits = tf.matmul(layer4_actv, variables['w5']) + variables['b5'] 
 
    #layer5_fccd = tf.matmul(layer4_actv, variables['w5']) + variables['b5'] 
    #layer5_actv = tf.nn.sigmoid(layer5_fccd) 
    #logits =tf.nn.sigmoid(tf.matmul(layer5_actv, variables['w6']) + variables['b6']) 
    return logits 

非常感谢您的帮助!

请您参考如下方法:

  1. 要在最后一个全连接层中找到图像的特征表示,只需在 TensorFlow session 中运行该张量即可。

    with tf.Session() as sess: 
        sess.run(layer4_actv, feed_dict={data: your_images_here}) 
    

    为了在 model_lenet5() 函数之外访问中间层的张量,您可能希望将每个张量存储在字典中并在最后返回它。

    def model_lenet5(data, variables): 
        model = {} 
        model['layer1_conv'] = tf.nn.conv2d(data, variables['w1'], [1, 1, 1, 1], padding='SAME') 
        model['layer1_actv'] = tf.sigmoid(layer1_conv + variables['b1']) 
        # etc... 
        return model 
     
     model = model_lenet5(data, variables) 
     with tf.Session() as sess: 
         sess.run(model['layer4_actv'], feed_dict={data: your_images_here})  
    

    或者,您可以使用 tf.get_default_graph().get_tensor_by_name() 找到与最后一个全连接层的输出对应的 Tensor。

    layer4_actv = tf.nn.sigmoid(layer4_fccd, name='layer4_actv') 
     
    with tf.Session() as sess: 
        sess.run(tf.get_default_graph().get_tensor_by_name('layer4_actv:0'), feed_dict={data: your_images_here}) 
    
  2. 对于张量 x,要将其形状作为列表,只需执行 x.shape.as_list()。您可以通过执行 flat_layer.shape.as_list()[0]

  3. 找到第一个全连接层的输入大小