Fix cross-GPU mask device mismatch in GSAM2 reconciliation

inference.py

@@ -1849,6 +1849,11 @@ def run_grounded_sam2_tracking(
             sam2_masks = MaskDictionary()
             tracking_results: Dict[int, Dict[int, ObjectInfo]] = {}
 
+            def _mask_to_cpu(mask):
+                if torch.is_tensor(mask):
+                    return mask.detach().cpu()
+                return mask
+
             for seg_idx in sorted(segment_data.keys()):
                 start_idx, mask_dict, segment_results = segment_data[seg_idx]
 
@@ -1862,7 +1867,7 @@ def run_grounded_sam2_tracking(
                                 {
                                     k: ObjectInfo(
                                         instance_id=v.instance_id,
-                                        mask=v.mask,
+                                        mask=_mask_to_cpu(v.mask),
                                         class_name=v.class_name,
                                         x1=v.x1, y1=v.y1,
                                         x2=v.x2, y2=v.y2,
@@ -1874,6 +1879,10 @@ def run_grounded_sam2_tracking(
                             )
                     continue
 
+                # Normalize keyframe masks to CPU before cross-GPU IoU matching.
+                for info in mask_dict.labels.values():
+                    info.mask = _mask_to_cpu(info.mask)
+
                 # IoU match + get local→global remapping
                 global_id_counter, remapping = (
                     mask_dict.update_masks_with_remapping(
@@ -1899,7 +1908,7 @@ def run_grounded_sam2_tracking(
                             continue
                         remapped[global_id] = ObjectInfo(
                             instance_id=global_id,
-                            mask=obj_info.mask,
+                            mask=_mask_to_cpu(obj_info.mask),
                             class_name=obj_info.class_name,
                             x1=obj_info.x1, y1=obj_info.y1,
                             x2=obj_info.x2, y2=obj_info.y2,

models/segmenters/grounded_sam2.py

@@ -164,13 +164,24 @@ class MaskDictionary:
 
     @staticmethod
     def _iou(m1: torch.Tensor, m2: torch.Tensor) -> float:
+        if not torch.is_tensor(m1):
+            m1 = torch.as_tensor(m1)
+        if not torch.is_tensor(m2):
+            m2 = torch.as_tensor(m2)
+
+        # Multi-GPU reconciliation can compare masks produced on different
+        # devices; normalize both masks onto CPU before arithmetic.
+        if m1.device != m2.device:
+            m1 = m1.detach().to(device="cpu")
+            m2 = m2.detach().to(device="cpu")
+
         m1f = m1.to(torch.float32)
         m2f = m2.to(torch.float32)
         inter = (m1f * m2f).sum()
         union = m1f.sum() + m2f.sum() - inter
         if union == 0:
             return 0.0
-        return float(inter / union)
+        return float((inter / union).item())
 
 
 # ---------------------------------------------------------------------------