Loads and rectifies an image from file. Args: image_path (str): path to an image from the dataset. model (camera_model.CameraMode

Decode a binary Velodyne example (of the form '<timestamp>.bin') Args: example_path (AnyStr): Oxford Radar RobotCar Dataset binary Velodyn

Decode a raw Velodyne example. (of the form '<timestamp>.png') Args: example_path (AnyStr): Oxford Radar RobotCar Dataset raw Velodyne exa

velodyne

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

Convert raw Velodyne data (from load_velodyne_raw) into a pointcloud Args: ranges (np.ndarray): Raw Velodyne range readings inten

Multi-Scale Deformable Attention Module :param d_model hidden dimension :param n_levels number of feature levels

Initializes convolutional block Args: in_channels: int, Number of input channels out_channels: int, Number of

Initializes depth distribution network. Args: constructor: function, Model constructor feat_extract_layer: st

NMS according to center distance

Generate the output sparse features from the focal sparse conv. Args: x: [N, C], lidar sparse features

Refer to https://github.com/tianweiy/CenterPoint L1 regression loss Args: regr (batch x max_objects x dim) gt_regr (batch

Refer to https://github.com/tianweiy/CenterPoint L1 regression loss Args: regr (batch x max_objects x dim) gt_regr (batch

Run all_gather on arbitrary picklable data (not necessarily tensors) Args: data: any picklable object Returns: list[data]

Args: im: (H, W, C) [y, x] x: (N) y: (N) Returns:

Builds modules

Creates an SE3 transform from translation and Euler angles. Args: xyzrpy (list[float]): translation and Euler angles for transform. Must

Args: x: x.features (N, C1) out_channels: C2 Returns:

Combine the foreground and background sparse features together. Args: x_fore: [N1, C], foreground sparse feat

Returns global embedding (np.array)

1x1 convolution

Create image depth feature volume by multiplying image features with depth distributions Args: image_features: (N, C, H,