Detection IoU ignores predictions without ground truth

[sirtris]

🐛 Bug

Hi,
in torchmetrics.detection.IntersectionOverUnion cases where a bounding box for a "non-existing" class are predicted are ignored. They should be 0 for that class instead.
Here "non-existing" means that the class does generally exist, just not in that image/examle.

To Reproduce

Take the snippet from the documentation and change the prediction:

import torch
from torchmetrics.detection import IntersectionOverUnion
preds = [
   {
       "boxes": torch.tensor([
            [296.55, 93.96, 314.97, 152.79],
            [298.55, 98.96, 314.97, 151.79]]),
       "labels": torch.tensor([4, 6]),
   }
]
target = [
   {
       "boxes": torch.tensor([
              [300.00, 100.00, 315.00, 150.00],
              [300.00, 100.00, 315.00, 150.00]
       ]),
       "labels": torch.tensor([4, 5]),
   }
]
metric = IntersectionOverUnion(class_metrics=True)
metric(preds, target)

The output is:
{'iou': tensor(0.6898), 'iou/cl_4': tensor(0.6898), 'iou/cl_5': tensor(nan)}

Expected behavior

First of all iou/cl_5 should be 0.0 instead of nan (#2778).
However I think there should also be an item with 'iou/cl_6': 0.0.

Environment

• TorchMetrics version: 1.6.1
• Python: 3.11.10
• PyTorch Version: 2.5.1