Python faster_rcnn_meta_arch.FasterRCNNMetaArch方法代码示例

# 需要导入模块: from object_detection.meta_architectures import faster_rcnn_meta_arch [as 别名]
# 或者: from object_detection.meta_architectures.faster_rcnn_meta_arch import FasterRCNNMetaArch [as 别名]
def test_create_faster_rcnn_models_from_config(
      self, use_matmul_crop_and_resize, enable_mask_prediction):
    model_proto = self.create_default_faster_rcnn_model_proto()
    faster_rcnn_config = model_proto.faster_rcnn
    faster_rcnn_config.use_matmul_crop_and_resize = use_matmul_crop_and_resize
    if enable_mask_prediction:
      faster_rcnn_config.second_stage_mask_prediction_loss_weight = 3.0
      mask_predictor_config = (
          faster_rcnn_config.second_stage_box_predictor.mask_rcnn_box_predictor)
      mask_predictor_config.predict_instance_masks = True

    for extractor_type, extractor_class in (
        model_builder.FASTER_RCNN_FEATURE_EXTRACTOR_CLASS_MAP.items()):
      faster_rcnn_config.feature_extractor.type = extractor_type
      model = model_builder.build(model_proto, is_training=True)
      self.assertIsInstance(model, faster_rcnn_meta_arch.FasterRCNNMetaArch)
      self.assertIsInstance(model._feature_extractor, extractor_class)
      if enable_mask_prediction:
        self.assertAlmostEqual(model._second_stage_mask_loss_weight, 3.0) 

# 需要导入模块: from object_detection.meta_architectures import faster_rcnn_meta_arch [as 别名]
# 或者: from object_detection.meta_architectures.faster_rcnn_meta_arch import FasterRCNNMetaArch [as 别名]
def test_create_faster_rcnn_models_from_config(self,
                                                 use_matmul_crop_and_resize,
                                                 enable_mask_prediction):
    model_proto = self.create_default_faster_rcnn_model_proto()
    faster_rcnn_config = model_proto.faster_rcnn
    faster_rcnn_config.use_matmul_crop_and_resize = use_matmul_crop_and_resize
    if enable_mask_prediction:
      faster_rcnn_config.second_stage_mask_prediction_loss_weight = 3.0
      mask_predictor_config = (
          faster_rcnn_config.second_stage_box_predictor.mask_rcnn_box_predictor)
      mask_predictor_config.predict_instance_masks = True

    for extractor_type, extractor_class in (
        self.faster_rcnn_feature_extractors().items()):
      faster_rcnn_config.feature_extractor.type = extractor_type
      model = model_builder.build(model_proto, is_training=True)
      self.assertIsInstance(model, faster_rcnn_meta_arch.FasterRCNNMetaArch)
      self.assertIsInstance(model._feature_extractor, extractor_class)
      if enable_mask_prediction:
        self.assertAlmostEqual(model._second_stage_mask_loss_weight, 3.0) 

# 需要导入模块: from object_detection.meta_architectures import faster_rcnn_meta_arch [as 别名]
# 或者: from object_detection.meta_architectures.faster_rcnn_meta_arch import FasterRCNNMetaArch [as 别名]
def _get_model(self, box_predictor, **common_kwargs):
    return faster_rcnn_meta_arch.FasterRCNNMetaArch(
        initial_crop_size=3,
        maxpool_kernel_size=1,
        maxpool_stride=1,
        second_stage_mask_rcnn_box_predictor=box_predictor,
        **common_kwargs) 

# 需要导入模块: from object_detection.meta_architectures import faster_rcnn_meta_arch [as 别名]
# 或者: from object_detection.meta_architectures.faster_rcnn_meta_arch import FasterRCNNMetaArch [as 别名]
def _compute_second_stage_input_feature_maps(self, features_to_crop,
                                               proposal_boxes_normalized,
                                               context_features,
                                               valid_context_size):
    """Crops to a set of proposals from the feature map for a batch of images.

    This function overrides the one in the FasterRCNNMetaArch. Aside from
    cropping and resizing the feature maps, which is done in the parent class,
    it adds context attention features to the box features.

    Args:
      features_to_crop: A float32 Tensor with shape [batch_size, height, width,
        depth]
      proposal_boxes_normalized: A float32 Tensor with shape [batch_size,
        num_proposals, box_code_size] containing proposal boxes in normalized
        coordinates.
      context_features: A float Tensor of shape [batch_size, context_size,
        num_context_features].
      valid_context_size: A int32 Tensor of shape [batch_size].

    Returns:
      A float32 Tensor with shape [K, new_height, new_width, depth].
    """
    box_features = self._crop_and_resize_fn(
        features_to_crop, proposal_boxes_normalized,
        [self._initial_crop_size, self._initial_crop_size])

    attention_features = self._context_feature_extract_fn(
        box_features=box_features,
        context_features=context_features,
        valid_context_size=valid_context_size)

    # Adds box features with attention features.
    box_features += attention_features

    flattened_feature_maps = self._flatten_first_two_dimensions(box_features)

    return self._maxpool_layer(flattened_feature_maps) 

# 需要导入模块: from object_detection.meta_architectures import faster_rcnn_meta_arch [as 别名]
# 或者: from object_detection.meta_architectures.faster_rcnn_meta_arch import FasterRCNNMetaArch [as 别名]
def _get_model(self, box_predictor, keras_model=False, **common_kwargs):
    return faster_rcnn_meta_arch.FasterRCNNMetaArch(
        initial_crop_size=3,
        maxpool_kernel_size=1,
        maxpool_stride=1,
        second_stage_mask_rcnn_box_predictor=box_predictor,
        **common_kwargs)