Fix in-place modification of user-input in SAM2 embed boxes (huggingface#42173)

* Do not modify boxes tensor in-place

Author: xenova

src/transformers/models/edgetam/modeling_edgetam.py

@@ -600,7 +600,7 @@ def _embed_points(self, points: torch.Tensor, labels: torch.Tensor, pad: bool) -
 
     def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
         """Embeds box prompts."""
-        boxes += 0.5  # Shift to center of pixel
+        boxes = boxes + 0.5  # Shift to center of pixel
         coords = boxes.view(*boxes.shape[:2], 2, 2)
         # add padding point for consistency with the original implementation
         coords = torch.nn.functional.pad(coords, (0, 0, 0, 1), mode="constant", value=0)

src/transformers/models/edgetam_video/modeling_edgetam_video.py

@@ -1644,7 +1644,7 @@ def _embed_points(self, points: torch.Tensor, labels: torch.Tensor, pad: bool) -
 
     def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
         """Embeds box prompts."""
-        boxes += 0.5  # Shift to center of pixel
+        boxes = boxes + 0.5  # Shift to center of pixel
         coords = boxes.view(*boxes.shape[:2], 2, 2)
         # add padding point for consistency with the original implementation
         coords = torch.nn.functional.pad(coords, (0, 0, 0, 1), mode="constant", value=0)

src/transformers/models/sam2/modeling_sam2.py

@@ -792,7 +792,7 @@ def _embed_points(self, points: torch.Tensor, labels: torch.Tensor, pad: bool) -
 
     def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
         """Embeds box prompts."""
-        boxes += 0.5  # Shift to center of pixel
+        boxes = boxes + 0.5  # Shift to center of pixel
         coords = boxes.view(*boxes.shape[:2], 2, 2)
         # add padding point for consistency with the original implementation
         coords = torch.nn.functional.pad(coords, (0, 0, 0, 1), mode="constant", value=0)

src/transformers/models/sam2/modular_sam2.py

@@ -890,7 +890,7 @@ def _embed_points(self, points: torch.Tensor, labels: torch.Tensor, pad: bool) -
 
     def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
         """Embeds box prompts."""
-        boxes += 0.5  # Shift to center of pixel
+        boxes = boxes + 0.5  # Shift to center of pixel
         coords = boxes.view(*boxes.shape[:2], 2, 2)
         # add padding point for consistency with the original implementation
         coords = torch.nn.functional.pad(coords, (0, 0, 0, 1), mode="constant", value=0)

src/transformers/models/sam2_video/modeling_sam2_video.py

@@ -1227,7 +1227,7 @@ def _embed_points(self, points: torch.Tensor, labels: torch.Tensor, pad: bool) -
 
     def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
         """Embeds box prompts."""
-        boxes += 0.5  # Shift to center of pixel
+        boxes = boxes + 0.5  # Shift to center of pixel
         coords = boxes.view(*boxes.shape[:2], 2, 2)
         # add padding point for consistency with the original implementation
         coords = torch.nn.functional.pad(coords, (0, 0, 0, 1), mode="constant", value=0)