Python inputs.transform_input_data方法代码示例

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_correct_class_label_encodings(self):
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
        fields.InputDataFields.groundtruth_boxes:
            tf.constant(np.array([[0, 0, 1, 1], [.5, .5, 1, 1]], np.float32)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32))
    }
    num_classes = 3
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=_fake_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=num_classes)
    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))

    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_classes],
        [[0, 0, 1], [1, 0, 0]])
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_confidences],
        [[0, 0, 1], [1, 0, 0]]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_correct_class_label_encodings(self):
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
        fields.InputDataFields.groundtruth_boxes:
            tf.constant(np.array([[0, 0, 1, 1], [.5, .5, 1, 1]], np.float32)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32))
    }
    num_classes = 3
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=_fake_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=num_classes)
    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))

    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_classes],
        [[0, 0, 1], [1, 0, 0]]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_correct_class_label_encodings(self):
        tensor_dict = {
            fields.InputDataFields.image:
                tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
            fields.InputDataFields.groundtruth_boxes:
                tf.constant(
                    np.array([[0, 0, 1, 1], [.5, .5, 1, 1]], np.float32)),
            fields.InputDataFields.groundtruth_classes:
                tf.constant(np.array([3, 1], np.int32))
        }
        num_classes = 3
        input_transformation_fn = functools.partial(
            inputs.transform_input_data,
            model_preprocess_fn=_fake_model_preprocessor_fn,
            image_resizer_fn=_fake_image_resizer_fn,
            num_classes=num_classes)
        with self.test_session() as sess:
            transformed_inputs = sess.run(
                input_transformation_fn(tensor_dict=tensor_dict))

        self.assertAllClose(
            transformed_inputs[fields.InputDataFields.groundtruth_classes],
            [[0, 0, 1], [1, 0, 0]]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_combine_additional_channels_if_present(self):
    image = np.random.rand(4, 4, 3).astype(np.float32)
    additional_channels = np.random.rand(4, 4, 2).astype(np.float32)
    def graph_fn(image, additional_channels):
      tensor_dict = {
          fields.InputDataFields.image: image,
          fields.InputDataFields.image_additional_channels: additional_channels,
          fields.InputDataFields.groundtruth_classes:
              tf.constant([1, 1], tf.int32)
      }

      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=1)
      out_tensors = input_transformation_fn(tensor_dict=tensor_dict)
      return out_tensors[fields.InputDataFields.image]
    out_image = self.execute_cpu(graph_fn, [image, additional_channels])
    self.assertAllEqual(out_image.dtype, tf.float32)
    self.assertAllEqual(out_image.shape, [4, 4, 5])
    self.assertAllClose(out_image, np.concatenate((image, additional_channels),
                                                  axis=2)) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_use_multiclass_scores_when_present(self):
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image: tf.constant(np.random.rand(4, 4, 3).
                                                    astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]],
                                   np.float32)),
          fields.InputDataFields.multiclass_scores:
              tf.constant(np.array([0.2, 0.3, 0.5, 0.1, 0.6, 0.3], np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32))
      }

      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=3, use_multiclass_scores=True)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return transformed_inputs[fields.InputDataFields.groundtruth_classes]
    groundtruth_classes = self.execute_cpu(graph_fn, [])
    self.assertAllClose(
        np.array([[0.2, 0.3, 0.5], [0.1, 0.6, 0.3]], np.float32),
        groundtruth_classes) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_correct_class_label_encodings(self):
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[0, 0, 1, 1], [.5, .5, 1, 1]], np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_classes],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_confidences])
    (groundtruth_classes, groundtruth_confidences) = self.execute_cpu(graph_fn,
                                                                      [])
    self.assertAllClose(groundtruth_classes, [[0, 0, 1], [1, 0, 0]])
    self.assertAllClose(groundtruth_confidences, [[0, 0, 1], [1, 0, 0]]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_applies_model_preprocess_fn_to_image_tensor(self):
    np_image = np.random.randint(256, size=(4, 4, 3))
    def graph_fn(image):
      tensor_dict = {
          fields.InputDataFields.image: image,
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }

      def fake_model_preprocessor_fn(image):
        return (image / 255., tf.expand_dims(tf.shape(image)[1:], axis=0))

      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.image],
              transformed_inputs[fields.InputDataFields.true_image_shape])
    image, true_image_shape = self.execute_cpu(graph_fn, [np_image])
    self.assertAllClose(image, np_image / 255.)
    self.assertAllClose(true_image_shape, [4, 4, 3]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_negative_size_error(self):
    """Test that error is raised for negative size boxes."""

    def graph_fn():
      tensors = {
          fields.InputDataFields.image: tf.zeros((128, 128, 3)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant([1, 1], tf.int32),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant([[0.5, 0.5, 0.4, 0.5]], tf.float32)
      }
      tensors = inputs.transform_input_data(
          tensors, _fake_model_preprocessor_fn, _fake_image_resizer_fn,
          num_classes=10)
      return tensors[fields.InputDataFields.groundtruth_boxes]
    with self.assertRaises(tf.errors.InvalidArgumentError):
      self.execute_cpu(graph_fn, []) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_negative_size_no_assert(self):
    """Test that negative size boxes are filtered out without assert.

    This test simulates the behaviour when we run on TPU and Assert ops are
    not supported.
    """

    tensors = {
        fields.InputDataFields.image: tf.zeros((128, 128, 3)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant([1, 1], tf.int32),
        fields.InputDataFields.groundtruth_boxes:
            tf.constant([[0.5, 0.5, 0.4, 0.5], [0.5, 0.5, 0.6, 0.6]],
                        tf.float32)
    }

    with mock.patch.object(tf, 'Assert') as tf_assert:
      tf_assert.return_value = tf.no_op()
      tensors = inputs.transform_input_data(
          tensors, _fake_model_preprocessor_fn, _fake_image_resizer_fn,
          num_classes=10)

      self.assertAllClose(tensors[fields.InputDataFields.groundtruth_boxes],
                          [[0.5, 0.5, 0.6, 0.6]]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_combine_additional_channels_if_present(self):
    image = np.random.rand(4, 4, 3).astype(np.float32)
    additional_channels = np.random.rand(4, 4, 2).astype(np.float32)
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(image),
        fields.InputDataFields.image_additional_channels:
            tf.constant(additional_channels),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([1, 1], np.int32))
    }

    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=_fake_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=1)
    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))
    self.assertAllEqual(transformed_inputs[fields.InputDataFields.image].dtype,
                        tf.float32)
    self.assertAllEqual(transformed_inputs[fields.InputDataFields.image].shape,
                        [4, 4, 5])
    self.assertAllClose(transformed_inputs[fields.InputDataFields.image],
                        np.concatenate((image, additional_channels), axis=2)) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_correct_merged_boxes(self):
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
        fields.InputDataFields.groundtruth_boxes:
            tf.constant(np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]], np.float32)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32))
    }

    num_classes = 3
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=_fake_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=num_classes,
        merge_multiple_boxes=True)

    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_boxes],
        [[.5, .5, 1., 1.]])
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_classes],
        [[1, 0, 1]])
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_confidences],
        [[1, 0, 1]])
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.num_groundtruth_boxes],
        1) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_resized_masks(self):
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
        fields.InputDataFields.groundtruth_instance_masks:
            tf.constant(np.random.rand(2, 4, 4).astype(np.float32)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32)),
        fields.InputDataFields.original_image_spatial_shape:
            tf.constant(np.array([4, 4], np.int32))
    }

    def fake_image_resizer_fn(image, masks=None):
      resized_image = tf.image.resize_images(image, [8, 8])
      results = [resized_image]
      if masks is not None:
        resized_masks = tf.transpose(
            tf.image.resize_images(tf.transpose(masks, [1, 2, 0]), [8, 8]),
            [2, 0, 1])
        results.append(resized_masks)
      results.append(tf.shape(resized_image))
      return results

    num_classes = 3
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=_fake_model_preprocessor_fn,
        image_resizer_fn=fake_image_resizer_fn,
        num_classes=num_classes,
        retain_original_image=True)
    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))
    self.assertAllEqual(transformed_inputs[
        fields.InputDataFields.original_image].dtype, tf.uint8)
    self.assertAllEqual(transformed_inputs[
        fields.InputDataFields.original_image_spatial_shape], [4, 4])
    self.assertAllEqual(transformed_inputs[
        fields.InputDataFields.original_image].shape, [8, 8, 3])
    self.assertAllEqual(transformed_inputs[
        fields.InputDataFields.groundtruth_instance_masks].shape, [2, 8, 8]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_applies_model_preprocess_fn_to_image_tensor(self):
    np_image = np.random.randint(256, size=(4, 4, 3))
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np_image),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32))
    }

    def fake_model_preprocessor_fn(image):
      return (image / 255., tf.expand_dims(tf.shape(image)[1:], axis=0))

    num_classes = 3
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=fake_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=num_classes)

    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))
    self.assertAllClose(transformed_inputs[fields.InputDataFields.image],
                        np_image / 255.)
    self.assertAllClose(transformed_inputs[fields.InputDataFields.
                                           true_image_shape],
                        [4, 4, 3]) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_applies_data_augmentation_fn_before_model_preprocess_fn(self):
    np_image = np.random.randint(256, size=(4, 4, 3))
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np_image),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32))
    }

    def mul_two_model_preprocessor_fn(image):
      return (image * 2, tf.expand_dims(tf.shape(image)[1:], axis=0))

    def add_five_to_image_data_augmentation_fn(tensor_dict):
      tensor_dict[fields.InputDataFields.image] += 5
      return tensor_dict

    num_classes = 4
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=mul_two_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=num_classes,
        data_augmentation_fn=add_five_to_image_data_augmentation_fn)
    with self.test_session() as sess:
      augmented_tensor_dict = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))

    self.assertAllEqual(augmented_tensor_dict[fields.InputDataFields.image],
                        (np_image + 5) * 2) 

# 需要导入模块: from object_detection import inputs [as 别名]
# 或者: from object_detection.inputs import transform_input_data [as 别名]
def test_returns_correct_merged_boxes(self):
    tensor_dict = {
        fields.InputDataFields.image:
            tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
        fields.InputDataFields.groundtruth_boxes:
            tf.constant(np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]], np.float32)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant(np.array([3, 1], np.int32))
    }

    num_classes = 3
    input_transformation_fn = functools.partial(
        inputs.transform_input_data,
        model_preprocess_fn=_fake_model_preprocessor_fn,
        image_resizer_fn=_fake_image_resizer_fn,
        num_classes=num_classes,
        merge_multiple_boxes=True)

    with self.test_session() as sess:
      transformed_inputs = sess.run(
          input_transformation_fn(tensor_dict=tensor_dict))
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_boxes],
        [[.5, .5, 1., 1.]])
    self.assertAllClose(
        transformed_inputs[fields.InputDataFields.groundtruth_classes],
        [[1, 0, 1]])