我正在从一个已训练的模型中恢复权重，并尝试使用预训练模型的权重来初始化另一层的某些层。我使用 session.run 和 get_tensor_by_name 从预训练模型中获取权重值。我用这些权重来初始化 tf.Variable。以下是我的代码：
checkpoint_dir = "check_point_" #directory that contains .meta, .index, checkpoint filescheckpoint_file = tf.train.latest_checkpoint(checkpoint_dir)graph = tf.Graph()with graph.as_default(): sess = tf.Session() with sess.as_default(): # Load the saved meta graph and restore variables saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file)) saver.restore(sess, checkpoint_file) # load and save data from model #f = open('weights.txt', 'ab') var = tf.global_variables() tf.initialize_all_variables().run() for v in var: print(v.name, end="\t") print(v.shape) gn = graph.get_tensor_by_name sr = sess.run v1 = sr(gn('Variable:0')) v2 = sr(gn('Variable_1:0')) v3 = sr(gn('Variable_2:0')) v4 = sr(gn('Variable_3:0')) v5 = sr(gn('Variable_4:0')) v6 = sr(gn('Variable_5:0')) v7 = sr(gn('Variable_6:0')) v8 = sr(gn('Variable_7:0')) print(type(v8)) conv1 = sess.run(gn('Variable:0')) train_data_node = tf.placeholder(tf.float32, shape=(BATCH_SIZE, IMAGE_SIZE, IMAGE_SIZE, NUM_CHANNELS)) conv1_weights = tf.Variable(v1, dtype=tf.float32) conv1_biases = tf.Variable(v2, dtype=tf.float32) # conv2_weights = tf.Variable(gn('Variable_2:0')) conv2_weights = tf.Variable(tf.truncated_normal( [5, 5, 32, 64], stddev=0.1, seed=SEED, dtype=tf.float32)) conv2_biases = tf.Variable(v4, dtype=tf.float32) fc1_weights = tf.Variable(v5, dtype=tf.float32) fc1_biases = tf.Variable(v6, dtype=tf.float32) fc2_weights = tf.Variable(v7, dtype=tf.float32) fc2_biases = tf.Variable(v8, dtype=tf.float32) # fc2_biases = tf.Variable(tf.constant( # 0.1, shape=[NUM_LABELS], dtype=tf.float32)) conv1 = tf.nn.conv2d(train_data_node, conv1_weights, strides=[1, 1, 1, 1], padding='SAME') # Bias and rectified linear non-linearity. relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_biases)) pool1 = tf.nn.max_pool(relu1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') 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)) pool2 = tf.nn.max_pool(relu2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') pool_shape = pool2.get_shape().as_list() reshape = tf.reshape( pool2, [pool_shape[0], pool_shape[1] * pool_shape[2] * pool_shape[3]]) # Fully connected layer. Note that the '+' operation automatically # broadcasts the biases. hidden = tf.nn.relu(tf.matmul(reshape, fc1_weights) + fc1_biases) print(sr(conv1_weights)) out = tf.matmul(hidden, fc2_weights) + fc2_biases print(sess.run(out, feed_dict={train_data_node: numpy.random.randn(200, 28, 28, 1)}))我遇到了这个错误：'Attempting to use uninitialized value Variable_9 [[{{node _retval_Variable_9_0_0}} = _Retval[T=DT_FLOAT, index=0, _device="/job:localhost/replica:0/task:0/device:CPU:0"](Variable_9)]]' 我做错了什么？
回答：
1) 如果我必须猜测的话，问题出在你执行了
tf.initialize_all_variables().run()如果我没记错的话，如果你从一个saver实例加载，你不需要执行 initialize_all_variables。
2) 如果你执行了 initialize_all_variables 并且你定义了自己的自定义变量，你需要在定义所有变量之后调用 initialize_all_variables
