本文整理汇总了Python中object_detection.exporter.replace_variable_values_with_moving_averages方法的典型用法代码示例。如果您正苦于以下问题：Python exporter.replace_variable_values_with_moving_averages方法的具体用法？Python exporter.replace_variable_values_with_moving_averages怎么用？Python exporter.replace_variable_values_with_moving_averages使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类object_detection.exporter的用法示例。
在下文中一共展示了exporter.replace_variable_values_with_moving_averages方法的3个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: test_replace_variable_values_with_moving_averages
# 需要导入模块: from object_detection import exporter [as 别名]
# 或者: from object_detection.exporter import replace_variable_values_with_moving_averages [as 别名]
def test_replace_variable_values_with_moving_averages(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
new_checkpoint_prefix = os.path.join(tmp_dir, 'new.ckpt')
self._save_checkpoint_from_mock_model(trained_checkpoint_prefix,
use_moving_averages=True)
graph = tf.Graph()
with graph.as_default():
fake_model = FakeModel()
preprocessed_inputs, true_image_shapes = fake_model.preprocess(
tf.placeholder(dtype=tf.float32, shape=[None, None, None, 3]))
predictions = fake_model.predict(preprocessed_inputs, true_image_shapes)
fake_model.postprocess(predictions, true_image_shapes)
exporter.replace_variable_values_with_moving_averages(
graph, trained_checkpoint_prefix, new_checkpoint_prefix)
expected_variables = set(['conv2d/bias', 'conv2d/kernel'])
variables_in_old_ckpt = self._get_variables_in_checkpoint(
trained_checkpoint_prefix)
self.assertIn('conv2d/bias/ExponentialMovingAverage',
variables_in_old_ckpt)
self.assertIn('conv2d/kernel/ExponentialMovingAverage',
variables_in_old_ckpt)
variables_in_new_ckpt = self._get_variables_in_checkpoint(
new_checkpoint_prefix)
self.assertTrue(expected_variables.issubset(variables_in_new_ckpt))
self.assertNotIn('conv2d/bias/ExponentialMovingAverage',
variables_in_new_ckpt)
self.assertNotIn('conv2d/kernel/ExponentialMovingAverage',
variables_in_new_ckpt) 示例2: test_replace_variable_values_with_moving_averages
# 需要导入模块: from object_detection import exporter [as 别名]
# 或者: from object_detection.exporter import replace_variable_values_with_moving_averages [as 别名]
def test_replace_variable_values_with_moving_averages(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
new_checkpoint_prefix = os.path.join(tmp_dir, 'new.ckpt')
self._save_checkpoint_from_mock_model(trained_checkpoint_prefix,
use_moving_averages=True)
graph = tf.Graph()
with graph.as_default():
fake_model = FakeModel()
preprocessed_inputs = fake_model.preprocess(
tf.placeholder(dtype=tf.float32, shape=[None, None, None, 3]))
predictions = fake_model.predict(preprocessed_inputs)
fake_model.postprocess(predictions)
exporter.replace_variable_values_with_moving_averages(
graph, trained_checkpoint_prefix, new_checkpoint_prefix)
expected_variables = set(['conv2d/bias', 'conv2d/kernel'])
variables_in_old_ckpt = self._get_variables_in_checkpoint(
trained_checkpoint_prefix)
self.assertIn('conv2d/bias/ExponentialMovingAverage',
variables_in_old_ckpt)
self.assertIn('conv2d/kernel/ExponentialMovingAverage',
variables_in_old_ckpt)
variables_in_new_ckpt = self._get_variables_in_checkpoint(
new_checkpoint_prefix)
self.assertTrue(expected_variables.issubset(variables_in_new_ckpt))
self.assertNotIn('conv2d/bias/ExponentialMovingAverage',
variables_in_new_ckpt)
self.assertNotIn('conv2d/kernel/ExponentialMovingAverage',
variables_in_new_ckpt) 示例3: test_replace_variable_values_with_moving_averages
# 需要导入模块: from object_detection import exporter [as 别名]
# 或者: from object_detection.exporter import replace_variable_values_with_moving_averages [as 别名]
def test_replace_variable_values_with_moving_averages(self):
tmp_dir = self.get_temp_dir()
trained_checkpoint_prefix = os.path.join(tmp_dir, 'model.ckpt')
new_checkpoint_prefix = os.path.join(tmp_dir, 'new.ckpt')
self._save_checkpoint_from_mock_model(trained_checkpoint_prefix,
use_moving_averages=True)
graph = tf.Graph()
with graph.as_default():
fake_model = FakeModel()
preprocessed_inputs, true_image_shapes = fake_model.preprocess(
tf.placeholder(dtype=tf.float32, shape=[None, None, None, 3]))
predictions = fake_model.predict(
preprocessed_inputs, true_image_shapes)
fake_model.postprocess(predictions, true_image_shapes)
exporter.replace_variable_values_with_moving_averages(
graph, trained_checkpoint_prefix, new_checkpoint_prefix)
expected_variables = set(['conv2d/bias', 'conv2d/kernel'])
variables_in_old_ckpt = self._get_variables_in_checkpoint(
trained_checkpoint_prefix)
self.assertIn('conv2d/bias/ExponentialMovingAverage',
variables_in_old_ckpt)
self.assertIn('conv2d/kernel/ExponentialMovingAverage',
variables_in_old_ckpt)
variables_in_new_ckpt = self._get_variables_in_checkpoint(
new_checkpoint_prefix)
self.assertTrue(expected_variables.issubset(variables_in_new_ckpt))
self.assertNotIn('conv2d/bias/ExponentialMovingAverage',
variables_in_new_ckpt)
self.assertNotIn('conv2d/kernel/ExponentialMovingAverage',
variables_in_new_ckpt)