MeanAveragePrecision throw error in DDP mode

[dreaquil]

🐛 Bug

Hi :) ! I am training a detection model (maskrcnn) and it works fine with the MeanAveragePrecision metric in single gpu mode. However, when I activate DDP with 4 GPUs, it throws an error on the first validation phase.

I have investigated and there are more detection labels than there are detections. However, I have gone through and made sure this is never the case when calling update on the metric (I added an assert).

  File "/home/andrea/projects/my-project/src/callbacks/detection_metrics.py", line 77, in on_validation_epoch_end
    mean_ap = self.val_mean_ap.compute()
  File "/home/andrea/projects/my-project/.venv/lib/python3.8/site-packages/torchmetrics/metric.py", line 523, in wrapped_func
    value = compute(*args, **kwargs)
  File "/home/andrea/projects/my-project/.venv/lib/python3.8/site-packages/torchmetrics/detection/mean_ap.py", line 908, in compute
    precisions, recalls = self._calculate(classes)
  File "/home/andrea/projects/my-project/.venv/lib/python3.8/site-packages/torchmetrics/detection/mean_ap.py", line 728, in _calculate
    ious = {
  File "/home/andrea/projects/my-project/.venv/lib/python3.8/site-packages/torchmetrics/detection/mean_ap.py", line 729, in <dictcomp>
    (idx, class_id): self._compute_iou(idx, class_id, max_detections)
  File "/home/andrea/projects/my-project/.venv/lib/python3.8/site-packages/torchmetrics/detection/mean_ap.py", line 468, in _compute_iou
    det = [det[i] for i in det_label_mask]
  File "/home/andrea/projects/my-project/.venv/lib/python3.8/site-packages/torchmetrics/detection/mean_ap.py", line 468, in <listcomp>
    det = [det[i] for i in det_label_mask]
IndexError: tuple index out of range

Additionally, I stepped through each update here and can confirm that each detection length matches item['labels] and item['scores'].

To Reproduce

Steps to reproduce the behavior...

Code sample

Expected behavior

Environment

• TorchMetrics version (and how you installed TM, e.g. conda, pip, build from source): 0.9.2 with pip
• Python & PyTorch Version (e.g., 1.0): Python 3.8.10, Pytorch 1.12
• Any other relevant information such as OS (e.g., Linux): Linux

Additional context