Convenience function for accessing box coordinates. Returns: a tensor with shape [N, 4] representing box coordinates.

Computes area of boxes. Args: boxlist: BoxList holding N boxes scope: name scope. Returns: a tensor with shape [N] represe

Name scope. Must be defined by implementations. Returns: a string representing the name scope of the anchor generation ope

Builds a KittiDataset object using the provided configurations Args: base_cfg: a base dataset configuration use_defau

Retrieve the next `batch_size` samples from this data set. Args: batch_size: number of samples in the batch

Compute pairwise intersection areas between boxes. Args: boxlist1: BoxList holding N boxes boxlist2: BoxList holding M boxes sc

Determines global path drop decision based on given probabilities. Args: p_img: A tensor of float32, probability of keeping image

Generates a collection of bounding boxes to be used as anchors.

Accesses a box collection and associated fields. This function returns specified field with object; if no field is specified, it retu

Loads input-output data for a set of samples. Should only be called when a particular sample dict is required. Otherwise, sam

Create optimizer learning rate based on config. Args: learning_rate_config: A LearningRate proto message. global_summaries: A set

Gets the full file path to the anchors info Args: classes_name: name of classes ('Car', 'Pedestrian', 'Cyclist',

Splits indices between children Args: dataset: Dataset object num_children: Number of children to split samples between Retu

Spawns children to do work Args: all_child_pids: List of child pids are appended here, the parent process should use this lis

Returns subsampled minibatch. Args: indicator: boolean tensor of shape [N] whose True entries can be sampled.

Compute loss function. Args: prediction_tensor: A float tensor of shape [batch_size, num_anchors, num_classes] represent

Add field to box list. This method can be used to add related box data such as weights/labels, etc. Args: field: a

Builds flattened output layers Args: tensor_in: Input tensor num_final_classes: Final number of output classes, including

Builds flattened output layers Args: tensor_in: Input tensor num_final_classes: Final number of output classes, including

Creates a slice filter to take a slice of the point cloud between ground_offset_dist and offset_dist above the ground plane Args

Filters ground truth labels based on class, difficulty, and maximum occlusion Args: objects: A list of ground truth insta

Returns all non-box fields (i.e., everything not named 'boxes').

Preprocesses the given input. Args: tensor_in: A `Tensor` of shape=(batch_size, height, width, channel) represent

Evaluates network metrics once over all the validation samples. Args: checkpoint_to_restore: The directory of the checkpoint to r

Copies the extra fields of boxlist_to_copy_from to boxlist_to_copy_to. Args: boxlist_to_copy_to: BoxList to which extra fields are copied.

Calculates cross entropy loss for classification Args: model: network model cls_logits: predicted classification logits c

Returns a unique file path for a text file based on the dataset name, split, object class, and number of clusters. The file p

Saves the anchors info matrix to a file Args: classes_name (str): classes name, e.g. 'Car', 'Pedestrian',

Modified VGG for image feature extraction. Note: All the fully_connected layers have been transformed to conv2d layers and are

Calculates the box_3d centroid xz, l, w, and ry from the 4 points of a box_4c. To calculate length and width, points are projected onto the di

Calculates accuracy of predicted objectness/classification wrt to the labels Args: cls_pred: A numpy array containing the

Converts an object class type string into a integer index Args: class_str: the object type (e.g. 'Car', 'Pedestrian', or

To be overridden Creates a feed_dict that can be passed into a tensorflow session Returns: a dictionary with tensors as keys and num

Gather boxes from BoxList according to indices and return new BoxList. By default, `gather` returns boxes corresponding to the input index list,

Returns the predictions and scores stacked for saving to file. Args: predictions: A dictionary containing the model outputs.

Finds the evaluated checkpoints. Examines the evaluation average losses file to find the already evaluated checkpoints. Args

Compute rotated 3D bounding box coordinates. :param boxes: a numpy array of the form: [[ry, l, h, w, tx, ty, tz],...

Returns the proposals and scores stacked for saving to file. Args: predictions: A dictionary containing the model outputs.

Factorize the projection matrix P as P=K*[R;t] and enforce the sign of the focal length to be fsign. Keyword Arguments: ----------------

Load the sample names listed in this dataset's set file (e.g. train.txt, validation.txt) Args: data_split: override the s

Compute scalar loss tensors with respect to provided groundtruth. Calling this function requires that groundtruth tensors have been p

Evaluation function for evaluating all the existing checkpoints. This function just runs through all the existing checkpoints. Raises

Samples a mini batch based on anchor ious with ground truth Args: max_ious: a tensor of max ious with ground truth in

Subsample indicator vector. Given a boolean indicator vector with M elements set to `True`, the function assigns all but `num_samples

Returns the diagonal lengths of box_8c Args: boxes_3d: An tensor of shape (N x 3 x 8) of boxes in box_8c format. Returns:

Builds classification, offset, and angle vector losses. Args: model: network model prediction_dict: prediction dictionary Re

Builds classification, and offset losses. Args: model: network model prediction_dict: prediction dictionary Returns:

Calculates the list of anchor information in the format: N x 8 [max_gt_2d_iou, max_gt_3d_iou, (6 x offsets), class_index]

This filters an object by class. Args: obj: An instance of ground-truth Object Label Returns: True or False depending on w

Checks that dataset directory exists in the file system Raises: FileNotFoundError: if the dataset folder is missing

This filters an object by difficulty. Args: obj: An instance of ground-truth Object Label difficulty: An int defining

Checks if a mini batch file exists already Args: classes_name (str): classes name, e.g. 'Car', 'Pedestrian',

Method to be overriden by implementations. Args: prediction_tensor: a tensor representing predicted quantities target

Returns a dictionary of the losses sum for averaging.

Returns a dictionary of the losses sum for averaging.

Fills anchor placeholder inputs with corresponding data Args: anchors_info: anchor info from mini_batch_utils

Splits ground truth labels based on provided classes Args: obj_labels: ObjectLabel list for an image classes

To be overridden by implementations. Args: **params: parameters for anchor generation op Returns: boxes: a BoxLi

Calculates the smooth L1 combined offset and angle loss, normalized by the number of positives Args: model: network model

Calculates the smooth L1 combined offset and angle loss, normalized by the number of positives Args: model: network model

Gets the positive mask based on the ground truth box classifications Args: positive_selection: positive selection method (e.g

Subtracts the given means from each image channel. For example: means = [123.68, 116.779, 103.939] image = _mean_image_subtra

Parses anchor indices, offsets, and classes from a matrix Args: anchors_info: an np.ndarray in the form

Reads cluster information from a text file Args: dataset: Dataset, used to find the saved text file cls: cla

Sets up data directories.