在检查点 Tensorflow 中找不到键

我正在使用 Tensorflow v1.1，并且一直在尝试找出如何使用 EMA 权重进行推理，但无论我做什么，我都会不断收到错误

未找到：在检查点中未找到 W/ExponentialMovingAverage 键

即使当我循环并打印出所有tf.global_variables密钥存在

这是一个可复制的脚本，很大程度上改编自面网的 https://github.com/davidsandberg/facenet/blob/master/test/restore_test.py单元测试：

import tensorflow as tf
import numpy as np


tf.reset_default_graph()

# Create 100 phony x, y data points in NumPy, y = x * 0.1 + 0.3
x_data = np.random.rand(100).astype(np.float32)
y_data = x_data * 0.1 + 0.3

# Try to find values for W and b that compute y_data = W * x_data + b
# (We know that W should be 0.1 and b 0.3, but TensorFlow will
# figure that out for us.)
W = tf.Variable(tf.random_uniform([1], -1.0, 1.0), name='W')
b = tf.Variable(tf.zeros([1]), name='b')
y = W * x_data + b

# Minimize the mean squared errors.
loss = tf.reduce_mean(tf.square(y - y_data))
optimizer = tf.train.GradientDescentOptimizer(0.5)
opt_op = optimizer.minimize(loss)

# Track the moving averages of all trainable variables.
ema = tf.train.ExponentialMovingAverage(decay=0.9999)
variables = tf.trainable_variables()
print(variables)
averages_op = ema.apply(tf.trainable_variables())
with tf.control_dependencies([opt_op]):
    train_op = tf.group(averages_op)

# Before starting, initialize the variables.  We will 'run' this first.
init = tf.global_variables_initializer()

saver = tf.train.Saver(tf.trainable_variables())

# Launch the graph.
sess = tf.Session()
sess.run(init)

# Fit the line.
for _ in range(201):
    sess.run(train_op)

w_reference = sess.run('W/ExponentialMovingAverage:0')
b_reference = sess.run('b/ExponentialMovingAverage:0')

saver.save(sess, os.path.join("model_ex1"))

tf.reset_default_graph()

tf.train.import_meta_graph("model_ex1.meta")
sess = tf.Session()

print('------------------------------------------------------')
for var in tf.global_variables():
    print('all variables: ' + var.op.name)
for var in tf.trainable_variables():
    print('normal variable: ' + var.op.name)
for var in tf.moving_average_variables():
    print('ema variable: ' + var.op.name)
print('------------------------------------------------------')

mode = 1
restore_vars = {}
if mode == 0:
    ema = tf.train.ExponentialMovingAverage(1.0)
    for var in tf.trainable_variables():
        print('%s: %s' % (ema.average_name(var), var.op.name))
        restore_vars[ema.average_name(var)] = var
elif mode == 1:
    for var in tf.trainable_variables():
        ema_name = var.op.name + '/ExponentialMovingAverage'
        print('%s: %s' % (ema_name, var.op.name))
        restore_vars[ema_name] = var

saver = tf.train.Saver(restore_vars, name='ema_restore')

saver.restore(sess, os.path.join("model_ex1")) # error happens here!

w_restored = sess.run('W:0')
b_restored = sess.run('b:0')

print(w_reference)
print(w_restored)
print(b_reference)
print(b_restored)

The key not found in checkpoint错误意味着该变量存在于内存中的模型中，但不存在于磁盘上的序列化检查点文件中。

您应该使用检查检查点工具 https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/inspect_checkpoint.py了解检查点中保存了哪些张量，以及为什么某些指数移动平均线没有保存在此处。

从您的重现示例中不清楚哪一行应该触发错误