Load state_dict to a module. This method is modified from :meth:`torch.nn.Module.load_state_dict`. Default value for ``strict`` is set to ``Fa

In distributed setting, this function only download checkpoint at local rank 0.

Find boxes that are non-empty. A box is considered empty, if either of its side is no larger than threshold. Returns:

Prepare data for testing. Args: index (int): Index for accessing the target data. Returns: dict: Testing dat

Args: rgb (bool): whether to return RGB colors or BGR colors. maximum (int): either 255 or 1 Returns: ndarray: a vec

Args: img: same as in __init__

Makes a function retry itself after encountering pytorch's CUDA OOM error. It will first retry after calling `torch.cuda.empty_cache()`.

Args: sample_points: 3D sampling points in shape [B, Q, T, G, P, 3] mlvl_feats: list of multi-scale features from neck, each in s

Args: sample_points: 3D sampling points in shape [B, Q, T, G, P, 3] mlvl_feats: list of multi-scale features from neck, each in s

Visualize the points cloud. Args: save_path (str, optional): path to save image. Default: None.

Hooks are stateless by default, but can be made checkpointable by implementing `state_dict` and `load_state_dict`.

Apply hard voxelization to points.

Args: stem (nn.Module): a stem module stages (list[list[CNNBlockBase]]): several (typically 4) stages,

Generate targets of line primitive. Args: point_mask (torch.Tensor): Tensor to store the ground truth of mask.

Create a grayscale version of the original image. The colors in masked area, if given, will be kept.

Compute loss for dense multi-level box regression. Loss is accumulated over ``fg_mask``. Args: anchors: #lvl anchor boxes, each

Paste instance masks according to boxes. This implementation is modified from https://github.com/facebookresearch/detectron2/ Args:

Draw points on visualizer. Args: points (numpy.array | torch.tensor, shape=[N, 3+C]): points to visualize. vis (:obj:

Draw points on visualizer. Args: points (numpy.array | torch.tensor, shape=[N, 3+C]): points to visualize. vis (:obj:

Repeat x `num` times to form a list.

Given feats and indexes, returns the gathered feats. Args: feats (torch.Tensor): Features to be transposed and gathered

Return an activation function given a string

Mask head forward function used in both training and testing.

Wrapper method that renders individual layers on an axis. :param ax: The matplotlib axes where the layer will get rendered. :

Transpose the dense per-level predictions. Args: predictions: a list of outputs, each is a list of per-level

Write points into ``obj`` format for meshlab visualization. Args: points (np.ndarray): Points in shape (N, dim). out_filename (st

Convert the rotation difference to difference in sine function. Args: boxes1 (torch.Tensor): Original Boxes in shape (NxC), where

Test function with augmentaiton.

Convert a list of bounding boxes to roi format. Args: bbox_list (list[torch.Tensor]): A list of bounding boxes corresponding

Multi-class NMS for 3D boxes. The IoU used for NMS is defined as the 2D IoU between BEV boxes. Args: mlvl_bboxes (torch.Tensor): Mult

Built an anchor generator from `cfg.MODEL.ANCHOR_GENERATOR.NAME`.

Build the whole model architecture, defined by ``cfg.MODEL.META_ARCHITECTURE``. Note that it does not load any weights from ``cfg``.

Build an optimizer from config.

Args: x_data (list[int]): a list containing x values of all the points being drawn. Length of list should match t

Extract features from images and points.

Extract features from images and points. Return: (BEV Feature, None, depth)

Extract features from images and points. Return: (BEV Feature, None, depth)

Directly extract features from the img backbone+neck.

Flatten an object so it can be used for PyTorch tracing. Also returns how to rebuild the original object from the flattened outputs. Ret

Forward function for training. Args: points (list[torch.Tensor]): List of points of shape [N, C]. img_metas (list): I

Generalized IoU from https://giou.stanford.edu/ The boxes should be in [x0, y0, x1, y1] format Returns a [N, M] pairwise matrix, where

Get anchors according to feature map sizes. Args: featmap_sizes (list[tuple]): Multi-level feature map sizes. input_m

Returns: dict: a dict which maps names (str) to data of the fields Modifying the returned dict will modify this instance

Returns: ndarray: the visualized image of shape (H, W, 3) (RGB) in uint8 type. The shape is scale

Args: win_inds indicates which windows a voxel belongs to. Voxels share a window have same inds. shape = [N,] Return:

Create boolean mask by actually number of a padded tensor. Args: actual_num ([type]): [description] max_num ([type]): [description

Create boolean mask by actually number of a padded tensor. Args: actual_num ([type]): [description] max_num ([type]): [description

Convert patch_box to shapely Polygon coordinates. :param patch_box: Patch box defined as [x_center, y_center, height, width].

Get points according to feature map sizes. Args: featmap_sizes (list[tuple]): Multi-level feature map sizes. dtype (t

Get the class of points according to coordinate type. Args: points_type (str): The type of points coordinate. The valid value

Compute the output size given input size and target long side length.

Get relative positional embeddings according to the relative positions of query and key sizes. Args: q_size (int): size of qu

Generate grid anchors in multiple feature levels. Args: featmap_sizes (list[tuple]): List of feature map sizes in

https://github.com/Megvii-BaseDetection/BEVStereo/blob/master/dataset/nusc_mv_det_dataset.py#L48