Class to compute similarity based on Intersection over Union (IOU) metric.



View source

Compute pairwise IOU similarity between ground truth boxes and anchors.

B: batch_size N: Number of groundtruth boxes. M: Number of anchor boxes.

boxes_1 a float Tensor with M or B * M boxes.
boxes_2 a float Tensor with N or B * N boxes, the rank must be less than or equal to rank of boxes_1.
boxes_1_masks a boolean Tensor with M or B * M boxes. Optional.
boxes_2_masks a boolean Tensor with N or B * N boxes. Optional.

A Tensor with shape [M, N] or [B, M, N] representing pairwise iou scores, anchor per row and groundtruth_box per colulmn.

Input shape:

  • boxes_1: [N, 4], or [B, N, 4]
  • boxes_2: [M, 4], or [B, M, 4]
  • boxes_1_masks: [N, 1], or [B, N, 1]
  • boxes_2_masks: [M, 1], or [B, M, 1]

Output shape:

[M, N], or [B, M, N]