Sets up feature extractors and stores feature maps and bottlenecks as member variables.

Sets up input placeholders by adding them to self._placeholders. Keys are defined as self.PL_*.

Helper function to calculate the evaluation average losses. Note: This function evaluates only classification and regression/offsets

Helper function to calculate the evaluation average losses. Note: This function evaluates classification, regression/offsets and

Helper function to calculate the evaluation average losses. Args: eval_rpn_losses: A dictionary containing all the average

Writes cluster information to a text file Args: file_path: path to text file clusters: clusters to write

Adds an image summary showing tiled feature maps Args: feature_maps: a tensor of feature maps to show, dimensions should be

Adds a multiple of the principle components, with magnitude from a Gaussian distribution with mean 0 and stdev 0.1 Args: img_data: O

Finds the min/max of each corner to align irregular corners. In the case where the regressed corners might be skewed, it tries to align each

Calculates and returns PCA of a set of images Args: image_set: List of images read with cv2.imread in np.uint8 format Returns:

Fills in the placeholders with the actual input values. Currently, only a batch size of 1 is supported Args: sample_

Returns all fields.

Updates the calculated accuracies for rpn and avod losses. Args: predictions: A dictionary containing the model outputs.

Calculates clusters for each class Returns: all_clusters: list of clusters for each class all_std_devs: list

Computes the intersection of the bases of oriented 3D bounding "box" and a set boxes of oriented 3D bounding boxes "boxes". :param box: a nu

Compute 3D height intersection and union between a box and a list of boxes :param box: a numpy array of the form: [ry, l, h, w, tx, ty, tz]

Computes height and width of boxes in boxlist. Args: boxlist: BoxList holding N boxes scope: name scope. Returns: Height:

initialize class names

initialize class names

initialize class names

Computes pairwise intersection-over-area between box collections. intersection-over-area (IOA) between two boxes box1 and box2 is defined as

Converts a set of network predictions into text files required for KITTI evaluation.

Plots AP scores from the native eval script and prints top 5 checkpoints for each metric

Generates anchors info which is used for mini batch sampling. Processing on 'Cars' can be split into multiple processes, see the Options sect

Samples a mini batch based on anchor ious with ground truth Args: mb_pos_mask: a boolean tensor mask of size [N] of posi

Compute intersection areas between corresponding boxes in two boxlists. Args: boxlist1: BoxList holding N boxes boxlist2: BoxList hol

Preprocesses anchor info and saves info to files Args: indices (int array): sample indices to process. If None, p

Generates rpn mini batch info for the kitti dataset Preprocesses data and saves data to files. Each file contains information

Projects 3D points to 2D points in image space. Args: points_3d: a list of float32 tensor of shape [3, None] cam_p2_matrix: a flo

Periodically evaluates the checkpoints inside the `checkpoint_dir`. This function evaluates all the existing checkpoints as they are being

Calls the kitti native C++ evaluation code. It first saves the predictions in kitti format. It then creates two child processes to ru

Samples a mini batch to train AVOD with preconfigured mini batch size and 2D iou ranges Args: anchor_ious: a tensor

Samples a mini batch to train the RPN with preconfigured mini batch size and 3D iou ranges Args: anchor_ious: a tens

Helper function to save the AVOD loss evaluation results. Args: eval_avod_losses: A dictionary containing the loss sums

Helper function to save the RPN loss evaluation results.

scale box coordinates in x and y dimensions. Args: boxlist: BoxList holding N boxes y_scale: (float) scalar tensor x_scale: (fl

Convenience function for setting box coordinates. Args: boxes: a tensor of shape [N, 4] representing box corners Raises:

Returns subsample of entries in indicator. Args: indicator: boolean tensor of shape [N] whose True entries can be

Tiles anchors over the area extents by using meshgrids to generate combinations of (x, y, z), (l, w, h) and ry. Args: area_exten

Defines the VGG arg scope. Args: weight_decay: The l2 regularization coefficient. Returns: An arg_scope.

Defines the VGG arg scope. Args: weight_decay: The l2 regularization coefficient. Returns: An arg_scope.

Defines the VGG arg scope. Args: weight_decay: The l2 regularization coefficient. Returns: An arg_scope.

Defines the VGG arg scope. Args: weight_decay: The l2 regularization coefficient. Returns: An arg_scope.

Call the loss function. Args: prediction_tensor: a tensor representing predicted quantities. target_tensor: a tensor

Compares the given object to the current ObjectLabel instance. :param other: object to compare to this instance against :return: True

Constructs a minibatch sampler.

Constructs box collection. Args: boxes: a tensor of shape [N, 4] representing box corners Raises: ValueError: if

Evaluator class for evaluating model's detection output. Args: model: An instance of DetectionModel dataset_config: D

Preprocesses anchors and saves info to files for RPN training Args: dataset: Dataset object mini_batch_dir: directory

Constructor. Args: model_config: configuration for the model

BEV maps created using slices of the point cloud. Args: config: bev_generator protobuf config kitti_utils: KittiUtils

Constructs a minibatch sampler. Args: positive_fraction: desired fraction of positive examples (scalar in [0,1]) Raises:

Args: model_config: configuration for the model train_val_test: "train", "val", or "test" dataset: the da

Args: model_config: configuration for the model train_val_test: "train", "val", or "test" dataset: the da