util.iou

lucid.models.objdet.util.iou(boxes_a: Tensor, boxes_b: Tensor) → Tensor

The iou function computes the Intersection over Union (IoU) between two sets of bounding boxes.

Function Signature

def iou(boxes_a: Tensor, boxes_b: Tensor) -> Tensor

Parameters

  • boxes_a (Tensor): A tensor of shape \((M, 4)\) where each row is a box in the format \((x_1, y_1, x_2, y_2)\).

  • boxes_b (Tensor): A tensor of shape \((N, 4)\) representing another set of boxes.

Returns

  • Tensor: A tensor of shape \((M, N)\) where the element at position \((i, j)\) contains the IoU between boxes_a[i] and boxes_b[j].

\[\text{IoU}(A, B) = \frac{\text{area}(A \cap B)}{\text{area}(A \cup B)}\]

Note

  • Boxes are expected in absolute coordinate format.

  • This function is differentiable, and gradients can be propagated if used in training pipelines.

Example

>>> from lucid.models.objdet.util import iou
>>> a = lucid.Tensor([[10, 10, 20, 20]])
>>> b = lucid.Tensor([[15, 15, 25, 25], [0, 0, 5, 5]])
>>> iou_matrix = iou(a, b)
>>> print(iou_matrix)
Tensor([[0.14285715, 0.        ]], ...)