Make copy of targets in GeneralizedRCNNTransform (#2227)

* Make copy of targets in GeneralizedRCNNTransform

* Fix flake8

test/test_models_detection_utils.py

+import copy
 import torch
 from torchvision.models.detection import _utils
+from torchvision.models.detection.transform import GeneralizedRCNNTransform
 import unittest
 from torchvision.models.detection import fasterrcnn_resnet50_fpn
 
@@ -33,6 +35,15 @@ class Tester(unittest.TestCase):
             # check that expected initial number of layers are frozen
             self.assertTrue(all(is_frozen[:exp_froz_params]))
 
+    def test_transform_copy_targets(self):
+        transform = GeneralizedRCNNTransform(300, 500, torch.zeros(3), torch.ones(3))
+        image = [torch.rand(3, 200, 300), torch.rand(3, 200, 200)]
+        targets = [{'boxes': torch.rand(3, 4)}, {'boxes': torch.rand(2, 4)}]
+        targets_copy = copy.deepcopy(targets)
+        out = transform(image, targets)  # noqa: F841
+        self.assertTrue(torch.equal(targets[0]['boxes'], targets_copy[0]['boxes']))
+        self.assertTrue(torch.equal(targets[1]['boxes'], targets_copy[1]['boxes']))
+
 
 if __name__ == '__main__':
     unittest.main()

torchvision/models/detection/transform.py

@@ -82,6 +82,18 @@ class GeneralizedRCNNTransform(nn.Module):
                 ):
         # type: (...) -> Tuple[ImageList, Optional[List[Dict[str, Tensor]]]]
         images = [img for img in images]
+        if targets is not None:
+            # make a copy of targets to avoid modifying it in-place
+            # once torchscript supports dict comprehension
+            # this can be simplified as as follows
+            # targets = [{k: v for k,v in t.items()} for t in targets]
+            targets_copy: List[Dict[str, Tensor]] = []
+            for t in targets:
+                data: Dict[str, Tensor] = {}
+                for k, v in t.items():
+                    data[k] = v
+                targets_copy.append(data)
+            targets = targets_copy
         for i in range(len(images)):
             image = images[i]
             target_index = targets[i] if targets is not None else None