Applies hard mining to anchorwise losses. Args: location_losses: Float tensor of shape [batch_size, num_anchors] representing ancho

Computes func(x, sel), with sel sampled from [0...num_cases-1]. If both preprocess_vars_cache AND key are the same between two calls, sel will be

Returns a key that can be used to index arg_scope dictionary.

Assert that correct number of anchors was generated. Args: anchors_list: A list of box_list.BoxList object holding anchors generated.

Assign groundtruth targets. Adds a background class to each one-hot encoding of groundtruth classes and uses target assigner to obtain regres

Builds a callable activation from config. Args: activation_fn: hyperparams_pb2.Hyperparams.activation Returns: Callable activation funct

Build a dictionary of batch_norm params from config. Args: batch_norm: hyperparams_pb2.ConvHyperparams.batch_norm proto. is_training: Wheth

Builds a classification loss based on the loss config. Args: loss_config: A losses_pb2.ClassificationLoss object. Returns: Loss based on

Build the detection graph.

Builds a faster_rcnn_meta_arch.FasterRCNNFeatureExtractor based on config. Args: feature_extractor_config: A FasterRcnnFeatureExtractor proto c

Builds a Faster R-CNN or R-FCN detection model based on the model config. Builds R-FCN model if the second_stage_box_predictor in the config is of

Build a tf initializer from config. Args: initializer: hyperparams_pb2.Hyperparams.regularizer proto. Returns: tf initializer. Raises

Builds a localization loss based on the loss config. Args: loss_config: A losses_pb2.LocalizationLoss object. Returns: Loss based on the

Constructs a NASNet image model.

Builds non-max suppresson based on the nms config. Args: nms_config: post_processing_pb2.PostProcessing.BatchNonMaxSuppression proto. Return

Expands tree hierarchy representation to parent-child dictionary. Args: hierarchy: labels hierarchy as JSON file. skip_root: if true skips ro

Constructs a PNASNet image model for proposal classifier features.

Builds a tf-slim regularizer from config. Args: regularizer: hyperparams_pb2.Hyperparams.regularizer proto. Returns: tf-slim regularizer

Builds score converter based on the config. Builds one of [tf.identity, tf.sigmoid, tf.softmax] score converters based on the config. Args:

Builds a ssd_meta_arch.SSDFeatureExtractor based on config. Args: feature_extractor_config: A SSDFeatureExtractor proto config from ssd.proto.

Builds an SSD detection model based on the model config. Args: ssd_config: A ssd.proto object containing the config for the desired SSDMe

Converts bbox center-size representation to corners representation. Args: centers: a tensor with shape [N, 2] representing bounding box centers

Computes clip window to use during post_processing. Computes a new clip window to use during post-processing based on `resized_image_shapes`

Compute CorLoc score for a single class. Args: detected_boxes: A numpy array of shape [N, 4] representing detected box coordinate

Method to be overridden by implementations. Args: prediction_tensor: a tensor representing predicted quantities target_tensor: a tens

Compute new dynamic shape for resize_to_range method.

Compute new static shape for resize_to_range method.

Labels as true/false positives detection tuples across all classes. Args: detected_box_tuples: A numpy array of structures with shape [N,],

Labels true/false positives of detections of an image across all classes. Args: detected_boxes: A float numpy array of shape [N, 4], repres

Labels boxes detected with the same class from the same image as tp/fp. Detection boxes are expected to be already sorted by score. Args:

Create classification targets for each anchor. Assign a classification target of for each anchor to the matching groundtruth label that is pr

Create classification weights for each anchor. Positive (matched) anchors are associated with a weight of positive_class_weight and negative

Constructs a SsdInceptionV3FeatureExtractor. Args: depth_multiplier: float depth multiplier for feature extractor pad_to_multiple: th

Constructs a SsdInceptionV2FeatureExtractor. Args: depth_multiplier: float depth multiplier for feature extractor pad_to_multiple: th

Returns a regression target for each anchor. Args: anchors: a BoxList representing N anchors groundtruth_boxes: a BoxList representin

Set regression weight for each anchor. Only positive anchors are set to contribute to the regression loss, so this method returns a weight of

Method to be overriden by implementations. Args: rel_codes: a tensor representing N relative-encoded boxes anchors: BoxList of anchor

Returns the default target for anchors to regress to. Default regression targets are set to zero (though in this implementation what these ta

Method to be overriden by implementations. Args: boxes: BoxList holding N boxes to be encoded anchors: BoxList of N anchors Retu

Export helper.

Extracts second stage box classifier features, to be overridden.

Constructs tensorflow detection graph and returns output tensors. Args: model: model to perform predictions with. create_input_dict_fn: fun

Extracts first stage RPN features, to be overridden.

Extracts RPN features. This function extracts two feature maps: a feature map to be directly fed to a box predictor (to predict location and

Compute the expected filter scaling at given PNASNet cell start_cell_num. In the pnasnet.py code, filter_scaling starts at 1.0. We instead adapt

Flattens `K-d` tensor along batch dimension to be a `(K-1)-d` tensor. Converts `inputs` with shape [A, B, ..., depth] into a tensor of shape

Left-right flip the boxes. Args: boxes: rank 2 float32 tensor containing the bounding boxes -> [N, 4]. Boxes are in normalized form

Up-down flip the boxes. Args: boxes: rank 2 float32 tensor containing the bounding boxes -> [N, 4]. Boxes are in normalized form mea

Left-right flip masks. Args: masks: rank 3 float32 tensor with shape [num_instances, height, width] representing instance masks. Retur

Up-down flip masks. Args: masks: rank 3 float32 tensor with shape [num_instances, height, width] representing instance masks. Returns:

Gathers the masks that correspond to classes. Args: instance_masks: A 4-D float32 tensor with shape [K, num_classes, mask_height, m

To be overridden by implementations. Args: feature_map_shape_list: list of (height, width) pairs in the format [(height_0, width_0)

Returns an absolute path to TFRecord file.

Helper function to put all proto fields into a dictionary. For many preprocessing steps, there's an trivial 1-1 mapping from proto fields to func

Return list of spatial dimensions for each feature map in a list. Args: feature_maps: a list of tensors where the ith tensor has shape

Returns an absolute path to label map file.

Computes the depth of the mask predictor convolutions. Computes the depth of the mask predictor convolutions given feature channels and numbe

Computes overlaps and scores between detected and groudntruth boxes. Args: detected_boxes: A numpy array of shape [N, 4] representing detec

Computes overlaps and scores between detected and groudntruth masks. Args: detected_boxes: A numpy array of shape [N, 4] representing detec

Computes overlaps and scores between detected and groundtruth tuples. Both detections and groundtruth boxes have the same class tuples. Args

Returns shapes to pad dataset tensors to before batching. Args: dataset: tf.data.Dataset object. max_num_boxes: Max number of groundtruth b

Returns the value of a field named step_name from proto. Args: preprocessor_step_config: A preprocessor_pb2.PreprocessingStep object. step_

Takes a 1-D image batch shape tensor and converts it to a 2-D tensor. Example: If 1-D image batch shape tensor is [2, 300, 300, 3]. The corre

Determines whether the key starts with a generic config dictionary key.

Check whether data fullfills the format of N*[ymin, xmin, ymax, xmin]. Args: data: a numpy array of shape [N, 4] representing box coordinat

Computes scalar box classifier loss tensors. Uses self._detector_target_assigner to obtain regression and classification targets for the seco

Computes scalar RPN loss tensors. Uses self._proposal_target_assigner to obtain regression and classification targets for the first stage RPN

Method to be overridden by implementations. Args: similarity_matrix: Float tensor of shape [N, M] with pairwise similarity where hi

Creates indicator matrix of non-pad elements of padded batch proposals. Args: num_proposals: Tensor of type tf.int32 with shape [batch_size

Predicts boxes and class scores. Args: image_features: A float tensor of shape [batch_size, height, width, channels] containing fea

Performs mask prediction. Args: image_features: A float tensor of shape [batch_size, height, width, channels] containing features f

Adds box predictors to RPN feature map to predict proposals. Note resulting tensors will not have been postprocessed. Args: rpn_box_pr

Predicts the output tensors from second stage of Faster R-CNN. Args: rpn_box_encodings: 4-D float tensor of shape [batch_size, num_

Predicts non-box, non-class outputs using refined detections. For training, masks as predicted directly on the box_classifier_features, which

Extracts groundtruth data from detection_model and prepares it for eval. Args: detection_model: A `DetectionModel` object. class_agnostic:

Pre-processes boxes before adding them to the VRDDetectionEvaluation. Phrase detection and Relation detection subclasses re-implement this method

Pre-processes boxes before adding them to the VRDDetectionEvaluation. Phrase detection and Relation detection subclasses re-implement this method

Removes anchors that (partially) fall outside an image. Also removes associated box encodings and objectness predictions. Args: box_en

Removes entries with invalid boxes. A box is invalid if either its xmax is smaller than its xmin, or its ymax is smaller than its ymin.

Rotate boxes counter-clockwise by 90 degrees. Args: boxes: rank 2 float32 tensor containing the bounding boxes -> [N, 4]. Boxes are