Split the layers in `layer_groups` into batchnorm (`bn_types`) and non-batchnorm groups.

Get box informations in the unified normative coordinate system for a given frame

Get lidar in the unified normative coordinate system for a given frame The intensity is normalized to fit [0-1] range (pandaset inten

Args: aggregate_func: xyz: (N, 3) xyz_features: (N, C) xyz_bs_idxs: (N) new_xyz:

Computes area of boxes. Args: boxes: Numpy array with shape [N, 4] holding N boxes Returns: a numpy array with shape [N*1] r

Args: boxes3d: (N, 7 + C) [x, y, z, dx, dy, dz, heading] in lidar coordinate Returns: aligned_bev_boxes: (N, 4) [x1, y1, x2,

7 -------- 4 /| /| 6 -------- 5 . | | | | . 3 -------- 0 |/ |/ 2 -------- 1 A

Set config keys via list (e.g., from command line).

Instantiates a Box instance from encoded vector representation. :param data: [<float>: 16]. Output from encode. :return: <Box

Args: batch_dict: Returns:

Returns an array with an average precision per class. Args: gt: list of dictionaries in the format described below. predictions:

Args: pts_rect: img_shape: calib: Returns:

1. align: move to box center 2. scale 3. offset 4. pt inside box = 1

Create a new `OptimWrapper` from `self` with another `layer_groups` but the same hyper-parameters.

r""" Calculates the pairwise distance between X and Z D[b, i, j] = l2 distance X[b, i] and Z[b, j] Parameters --------- X : torch.Te

Conert pandaset pose dict to a numpy vector in order to pass it through the network

rotated box iou running in gpu. 500x faster than cpu version (take 5ms in one example with numba.cuda code). convert from [this project](

Args: rois: (M, 7 + C) points: (N, 3) sample_radius_with_roi: num_max_points_of_part: Returns: sampl

Return list of trainable params in `m`.

To support a custom dataset, implement this function to load the raw data (and labels), then transform them to the unified normative

Args: root_path: dataset_cfg: class_names: training: logger:

NMS according to center distance