Getting different values for IOU using the class vs the module

[MotivaoCrypto]

Hi all,

So I'm trying to generate metrics for IOU using the torchmetrics package and I realized I'm getting inconsistencies for the calculation or maybe I'm understanding this incorrectly. Either way, your help would be greatly appreciated.

Here is the python code for reproduction:

import torch



preds = [{
    "boxes": torch.tensor([[1497.,  395., 1733.,  574.],
        [1148.,  523., 1429.,  771.],
        [ 534.,  397.,  798.,  605.],
        [ 186.,  350.,  438.,  520.]]),
    "scores": torch.tensor([0.9840, 0.9910, 0.9830, 0.9930]),
    "labels": torch.tensor([1, 1, 1, 1], dtype=torch.int32)
}]

target =[{
    "labels": torch.tensor([1, 1, 1, 1], dtype=torch.int32),
    "iscrowd": torch.tensor([0, 0, 0, 0], dtype=torch.int32), 
    "area": torch.tensor([38127., 63468., 44718., 36498.]), 
    "boxes": torch.tensor([[1506.,  385., 1719.,  564.],
        [1163.,  512., 1421.,  758.],
        [ 535.,  418.,  792.,  592.],
        [ 195.,  360.,  432.,  514.]])
}]

pred_bbox = torch.tensor([[1497.,  395., 1733.,  574.],
        [1148.,  523., 1429.,  771.],
        [ 534.,  397.,  798.,  605.],
        [ 186.,  350.,  438.,  520.]], dtype=torch.float)
target_bbox = torch.tensor([[1506.,  385., 1719.,  564.],
        [1163.,  512., 1421.,  758.],
        [ 535.,  418.,  792.,  592.],
        [ 195.,  360.,  432.,  514.]], dtype=torch.float)
 

################## USING CLASS: #########################
from torchmetrics.detection import IntersectionOverUnion

iou_metric = IntersectionOverUnion()
iou_metric(preds, target)
print("Intersection Over Union:")
print(iou_metric.compute())
############################### USING MODULE:
from torchmetrics.functional.detection import intersection_over_union
print("IOU",intersection_over_union(target_bbox , pred_bbox ))

from torchvision.ops import box_iou
iou = box_iou(target_bbox , pred_bbox )
print("Box IOU:", iou)
print("Box IOU DIAG MEAN:", iou.diag().mean())

OUTPUT:

Intersection Over Union:
{'iou': tensor(0.2071)}
IOU tensor(0.8284)
Box IOU: tensor([[0.8112, 0.0000, 0.0000, 0.0000],
        [0.0000, 0.8360, 0.0000, 0.0000],
        [0.0000, 0.0000, 0.8144, 0.0000],
        [0.0000, 0.0000, 0.0000, 0.8520]])
Box IOU DIAG MEAN: tensor(0.8284))