Initializes directories, and loads the sample list Args: dataset_config: KittiDatasetConfig name: unique

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

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

Compute loss function. Args: prediction_tensor: A float tensor of shape [num_anchors, code_size] representing the

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

Reshapes a list of cluster data from a list of arrays into an N x 3 ndarray, for example a list of [2, 1, 2] clusters: [

Generates 3D anchors in a grid in the provided 3d area and places them on the ground_plane. Args: **params:

Calls add_feature_maps for a specified layer in a dictionary of end points Args: end_points: dictionary of network end points

Adds a single scalar summary value to the logs without adding a summary node to the graph Args: summary_name: name of the summar

Encodes the anchor regression predictions with the ground truth. Args: anchors: A numpy array of shape (N, 6) representing

Converts an anchor form [x, y, z, dim_x, dim_y, dim_z] to 3d box format of [x, y, z, l, w, h, ry] Note: In this conversion, if the flag 'fix_

Applies PCA jitter or random noise to a single image Args: image_in: Image to modify Returns: Modified image

Retrieves specified fields as a dictionary of tensors. Args: fields: (optional) list of fields to return in the dictionary.

Select boxes from BoxList according to indicator and return new BoxList. `boolean_mask` returns the subset of boxes that are marked as "True" by

measure overlap between 3D bounding boxes, parametrized by (ry, h, w, l, tx, ty, tz)

measure overlap between 3D bounding boxes, parametrized by (ry, h, w, l, tx, ty, tz)

measure overlap between 3D bounding boxes, parametrized by (ry, h, w, l, tx, ty, tz)

Returns a numpy array of 3d box format for iou calculation Args: boxes_3d: list of 3d boxes Returns: new_anchor_list: numpy a

Converts a box_3d [x, y, z, l, w, h, ry] into anchor form [x, y, z, dim_x, dim_y, dim_z] Anchors in box_3d format should have an ry of 0 or

Turns a box_3d into an ObjectLabel Args: box_3d: 3D box in the format [x, y, z, l, w, h, ry] obj_type: Optional, the object type

Computes the 3D bounding box corner positions from 8 corners. To go back from 8-corner representation to box3D, we need to reverse the transf

Builds basic layers Args: fc_layers_config: Fully connected layers config object input_rois: List of input roi feature m

Builds fusion layers Args: fc_layers_config: Fully connected layers config object input_rois: List of input roi feature maps

Builds the loss for a variety of box representations Args: model: network model prediction_dict: prediction dictionary Retur

Create optimizer based on config. Args: optimizer_config: A Optimizer proto message. global_summaries: A set to attach learning r

Builds second stage fully connected layers Args: layers_config: Configuration object input_rois: List of input ROI feature maps

Modified VGG for BEV feature extraction with pyramid features Args: inputs: a tensor of size [batch_size, height, width, channel

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

VGG 11-Layers modified version. Note: All the fully_connected layers have been transformed to conv2d layers. Args:

Modified VGG for image feature extraction with pyramid features. Args: inputs: a tensor of size [batch_size, height, width, chan

Converts between a list of objects and a numpy array containing the bounding boxes. :param obj_list: an object list as per object class

Calculates the point on the 3D plane for a point with one value missing :param plane: Coefficients of the plane equation (a, b, c, d) :param

Change coordinate frame of the boxlist to be relative to window's frame. Given a window of the form [ymin, xmin, ymax, xmax], changes boundin

Checks for correct anchor format. If not proper type, raises error. Args: input_data: input numpy array or tensor to check for valid anch

Checks for correct box_3d format. If not proper type, raises error. Args: input_data: input numpy array or tensor to check for valid box_

Checks for correct box_4c format. If not proper type, raises error. Args: input_data: input numpy array or tensor to check for valid box_

Checks for correct box_8c format. If not proper type, raises error. Args: input_data: input numpy array or tensor to check for valid box_

Computes the 3D bounding box corner positions from an ObjectLabel :param object_label: ObjectLabel to compute corners from :return: a numpy a

Computes the orientation given object and camera matrix Keyword arguments: obj -- object file to draw bounding box p -- transform matrix

Copies and compiles kitti native code. It also creates neccessary directories for storing the results of the kitti native evaluation code.

Checks if a directory exists, or else create it Args: dir: directory to create

Transform disparity map to 3d point cloud. Camera coordinate frame: X: right Y: down Z: forward Example Usage found in:

Calculates the signed distance from a 3D plane to each point in a list of points :param plane: Coefficients of the plane equation (a, b, c, d

Applies feature fusion to multiple inputs Args: fusion_method: 'mean' or 'concat' inputs: Input tensors of shape (batch_size, wid

Flips boxes_3d Args: boxes_3d: List of boxes in box_3d format flip_ry bool: (optional) if False, rotation is not flipped to save

Flips the ground plane by negating the x coefficient (ax + by + cz + d = 0) Args: ground_plane: ground plane coefficients Re

Flips an image horizontally

Flips only the 3D position of an object label. The 2D bounding box is not flipped to save time since it is not used. Args: obj_label:

Flips a point cloud (3, N)

Flips a list of points (N, 3)

Flips the stereo calibration matrix to correct the projection back to image space. Flipping the image can be seen as a movement of both the ca

Generates the BEV maps dictionary. One height map is created for each slice of the point cloud. One density map is created for the who

Generates BEV maps Args: **params: additional keyword arguments for specific implementations of BevGenerator.

Reads in the file containing the information matrix Args: classes_name: object type, one of ('Car', 'Pedestrian',

Computes the center coordinates, height and width of the boxes. Args: scope: name scope of the function. Returns:

Reads model configuration from a pipeline_pb2.NetworkPipelineConfig. Args: pipeline_config_path: A path directory to the network pipeline

Reads the ground plane for the sample Args: sample_name: name of the sample, e.g. '000123' Returns: ground_p

Calculates the lidar point cloud, and optionally returns only the points that are projected to the image. :param img_idx: image index :p

Reads model configuration from a configuration file. This merges the layer config info with model default configs. Args: config_pat

Creates a point filter using the 3D extents and ground plane :param point_cloud: Point cloud in the form [[x,...],[y,...],[z,...]] :para

Reads the road plane from file :param int img_idx : Index of image :param str planes_dir : directory containing plane text files :return

Extract parameters required to transform disparity image to 3D point cloud. Keyword Arguments: ------------------ left_cam_mat : 3x4

measure overlap between bird's eye view bounding boxes, parametrized by (ry, l, w, tx, tz)

measure overlap between bird's eye view bounding boxes, parametrized by (ry, l, w, tx, tz)

measure overlap between bird's eye view bounding boxes, parametrized by (ry, l, w, tx, tz)

criterion defines whether the overlap is computed with respect to both areas (ground truth and detection) or with respect to box a or b (detection and

criterion defines whether the overlap is computed with respect to both areas (ground truth and detection) or with respect to box a or b (detection and

criterion defines whether the overlap is computed with respect to both areas (ground truth and detection) or with respect to box a or b (detection and

Creates dense vector with indices set to specific value and rest to zeros. This function exists because it is unclear if it is safe to u

Computes pairwise intersection-over-union between box collections. Args: boxlist1: BoxList holding N boxes boxlist2: BoxList holding

Check if each point in a 3D point cloud lies within the 3D bounding box If we think of the bounding box as having bottom face defined by [P1,

Transforms the pointclouds to the camera 0 frame. Keyword Arguments: ------------------ xyz_lidar : N x 3 Numpy Array

Restores the model weights. Loads the weights loaded from checkpoint dir onto the model. It ignores the missing weights since this is used

Compute intersection-over-union between corresponding boxes in boxlists. Args: boxlist1: BoxList holding N boxes boxlist2: BoxList ho