Resolve MPS's lack of cummax

Author: sadra-barikbin

ignite/metrics/vision/object_detection_average_precision_recall.py

@@ -1,3 +1,4 @@
+import os
 from typing import Callable, cast, Dict, List, Optional, Sequence, Tuple, Union
 
 import torch
@@ -125,7 +126,7 @@ def box_iou(pred_boxes: torch.Tensor, gt_boxes: torch.Tensor, iscrowd: torch.Boo
             class_mean=None,
         )
         precision = torch.double if torch.device(device).type != "mps" else torch.float32
-        self.rec_thresholds = self.rec_thresholds.to(device=device, dtype=precision)
+        self.rec_thresholds = cast(torch.Tensor, self.rec_thresholds).to(device=device, dtype=precision)
 
     @reinit__is_reduced
     def reset(self) -> None:
@@ -234,7 +235,10 @@ def _compute_average_precision(self, recall: torch.Tensor, precision: torch.Tens
         Returns:
             average_precision: (n-1)-dimensional tensor containing the average precision for mean dimensions.
         """
+        mps_cpu_fallback = os.environ.get("PYTORCH_ENABLE_MPS_FALLBACK", "0")
+        os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
         precision_integrand = precision.flip(-1).cummax(dim=-1).values.flip(-1)
+        os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = mps_cpu_fallback
         rec_thresholds = cast(torch.Tensor, self.rec_thresholds).repeat((*recall.shape[:-1], 1))
         rec_thresh_indices = (
             torch.searchsorted(recall, rec_thresholds)