python类conv2d_transpose()的实例源码

def testOutputSizeWithStrideOneValidPadding(self):
    num_filters = 32
    input_size = [5, 10, 12, 3]
    expected_size = [5, 12, 14, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [3, 3], stride=1, padding='VALID')
    self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')

    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWithStrideTwoValidPadding(self):
    num_filters = 32
    input_size = [5, 9, 11, 3]
    expected_size = [5, 19, 23, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [3, 3], stride=[2, 2], padding='VALID')
    self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
    self.assertListEqual(list(output.get_shape().as_list()), expected_size)

    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWith1x1StrideTwoSamePadding(self):
    num_filters = 1
    input_size = [1, 1, 1, 1]
    expected_size = [1, 2, 2, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [2, 2], stride=[2, 2], padding='SAME')
    self.assertListEqual(list(output.get_shape().as_list()), expected_size)

    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWith1x1StrideTwoValidPadding(self):
    num_filters = 1
    input_size = [1, 1, 1, 1]
    expected_size = [1, 2, 2, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [2, 2], stride=[2, 2], padding='VALID')
    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWith2x2StrideTwoSamePadding(self):
    num_filters = 1
    input_size = [1, 2, 2, 1]
    expected_size = [1, 4, 4, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [2, 2], stride=[2, 2], padding='SAME')
    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWithStride2x1(self):
    num_filters = 1
    input_size = [1, 3, 2, 1]
    expected_size = [1, 6, 5, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [2, 4], stride=[2, 1], padding='VALID')
    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWithStride2x4(self):
    num_filters = 1
    input_size = [1, 3, 2, 1]
    expected_size = [1, 6, 8, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [2, 4], stride=[2, 4], padding='VALID')
    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeWithStride2x5(self):
    num_filters = 1
    input_size = [1, 3, 2, 1]
    expected_size = [1, 6, 10, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [2, 4], stride=[2, 5], padding='VALID')
    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
      self.assertListEqual(list(output.eval().shape), expected_size)

def testOutputSizeRandomSizesAndStridesValidPadding(self):
    np.random.seed(0)
    max_image_size = 10

    for _ in range(10):
      num_filters = 1
      input_size = [
          1, np.random.randint(1, max_image_size),
          np.random.randint(1, max_image_size), 1
      ]
      filter_size = [
          np.random.randint(1, input_size[1] + 1),
          np.random.randint(1, input_size[2] + 1)
      ]
      stride = [np.random.randint(1, 3), np.random.randint(1, 3)]

      ops.reset_default_graph()
      graph = ops.Graph()
      with graph.as_default():
        images = random_ops.random_uniform(input_size, seed=1)
        transpose = layers_lib.conv2d_transpose(
            images, num_filters, filter_size, stride=stride, padding='VALID')
        conv = layers_lib.conv2d(
            transpose, num_filters, filter_size, stride=stride, padding='VALID')

        with self.test_session(graph=graph) as sess:
          sess.run(variables_lib.global_variables_initializer())
          self.assertListEqual(list(conv.eval().shape), input_size)

def testWithScope(self):
    num_filters = 32
    input_size = [5, 9, 11, 3]
    expected_size = [5, 19, 23, num_filters]

    images = random_ops.random_uniform(input_size, seed=1)
    output = layers_lib.conv2d_transpose(
        images, num_filters, [3, 3], stride=2, padding='VALID', scope='conv7')
    self.assertEqual(output.op.name, 'conv7/Relu')

    with self.test_session() as sess:
      sess.run(variables_lib.global_variables_initializer())
      self.assertListEqual(list(output.eval().shape), expected_size)

def testDeconvWithoutBiasesProducesConv2dTranspose(self):
    num_filters = 32
    input_size = [5, 9, 11, 3]
    expected_size = [5, 19, 23, num_filters]
    stride = 2
    padding = 'VALID'

    with self.test_session() as sess:
      images = random_ops.random_uniform(input_size, seed=1)
      output_deconv = layers_lib.conv2d_transpose(
          images,
          num_filters, [3, 3],
          stride=stride,
          padding=padding,
          activation_fn=None,
          scope='conv7')

      weights = variables.get_variables_by_name('conv7/weights')[0]
      output_conv2d_transpose = nn_ops.conv2d_transpose(
          images,
          weights,
          expected_size, [1, stride, stride, 1],
          padding=padding)

      sess.run(variables_lib.global_variables_initializer())

      output_deconv, output_conv2d_transpose = sess.run(
          [output_deconv, output_conv2d_transpose])

      self.assertTrue(
          np.isclose(output_deconv, output_conv2d_transpose, 1e-5, 1e-5).all())