vis4d.op.detect3d.util

Utilitiy functions for detection 3D ops.

Functions

batched_nms_rotated(boxes, scores, idxs, ...)

Performs non-maximum suppression in a batched fashion.

bev_3d_nms(center_x, center_y, width, ...[, ...])

BEV 3D NMS.
batched_nms_rotated(boxes, scores, idxs, iou_threshold)[source]

Performs non-maximum suppression in a batched fashion.

Each index value correspond to a category, and NMS will not be applied between elements of different categories.

Parameters:

  • boxes (Tensor) – Boxes where NMS will be performed. They are expected to be in (x_ctr, y_ctr, width, height, angle_degrees) format. In shape (N, 5).

  • scores (Tensor) – Scores for each one of the boxes. In shape (N,).

  • idxs (Tensor) – Indices of the categories for each one of the boxes. In shape (N,).

  • iou_threshold (float) – Discards all overlapping boxes with IoU < iou_threshold.

Returns:

Int64 tensor with the indices of the elements that have been

kept by NMS, sorted in decreasing order of scores

Return type:

Tensor

bev_3d_nms(center_x, center_y, width, length, angle, scores, class_ids=None, iou_threshold=0.1)[source]

BEV 3D NMS.

Parameters:

  • center_x (Tensor) – Center x of boxes. In shape (N, 1).

  • center_y (Tensor) – Center y of boxes. In shape (N, 1).

  • width (Tensor) – Width of boxes. In shape (N, 1).

  • length (Tensor) – Length of boxes. In shape (N, 1).

  • angle (Tensor) – Angle of boxes. In shape (N, 1).

  • scores (Tensor) – Scores of boxes. In shape (N, 1).

  • class_ids (Tensor | None, optional) – Class ids of boxes. In shape (N,). Defaults to None. If None, class_agnostic NMS will be performed.

  • iou_threshold (float, optional) – IoU threshold. Defaults to 0.1.

Returns:

Indices of boxes that have been kept by NMS.

Return type:

Tensor