From fb53aee50fb8d0b73a200334b4297ac445fd9ddf Mon Sep 17 00:00:00 2001
From: Rifa Ferzana <78103780+Rifa-111@users.noreply.github.com>
Date: Sun, 8 Nov 2026 00:47:34 +0000
Subject: [PATCH 1/2] Fix ShiftIntensity safe integer offset overflow

Promote integer tensors before adding offsets when safe=True to avoid
wraparound before dtype conversion, and cover signed/unsigned cases.

Signed-off-by: Rifa Ferzana <78103780+Rifa-111@users.noreply.github.com>
---
 monai/transforms/intensity/array.py      |  5 ++++-
 tests/transforms/test_shift_intensity.py | 20 ++++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index 0421d344923..f93c5f8d3e7 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -252,7 +252,10 @@ def __call__(self, img: NdarrayOrTensor, offset: float | None = None) -> Ndarray
 
         img = convert_to_tensor(img, track_meta=get_track_meta())
         offset = self.offset if offset is None else offset
-        out = img + offset
+        # Promote integer inputs before addition when safe=True. Otherwise integer
+        # offsets can overflow in the input dtype before safe conversion clamps.
+        computation_img = img.to(torch.float64) if self.safe and not img.is_floating_point() else img
+        out = computation_img + offset
         out, *_ = convert_data_type(data=out, dtype=img.dtype, safe=self.safe)
 
         return out
diff --git a/tests/transforms/test_shift_intensity.py b/tests/transforms/test_shift_intensity.py
index 70a6c7bf799..4b626fc5eb6 100644
--- a/tests/transforms/test_shift_intensity.py
+++ b/tests/transforms/test_shift_intensity.py
@@ -14,12 +14,32 @@
 import unittest
 
 import numpy as np
+import torch
 
 from monai.transforms import ShiftIntensity
 from tests.test_utils import NumpyImageTestCase2D
 
 
 class TestShiftIntensity(NumpyImageTestCase2D):
+    def test_safe_integer_offset_clamps_unsigned_image(self):
+        image = torch.tensor([0, 100, 255], dtype=torch.uint8)
+        for offset, expected in (
+            (1, [1, 101, 255]),
+            (-1, [0, 99, 254]),
+            (1.0, [1, 101, 255]),
+            (-1.0, [0, 99, 254]),
+        ):
+            with self.subTest(offset=offset):
+                result = ShiftIntensity(offset=offset, safe=True)(image)
+                torch.testing.assert_close(result, torch.tensor(expected, dtype=torch.uint8))
+
+    def test_safe_integer_offset_clamps_signed_image(self):
+        image = torch.tensor([-128, 0, 127], dtype=torch.int8)
+        result = ShiftIntensity(offset=1, safe=True)(image)
+        torch.testing.assert_close(result, torch.tensor([-127, 1, 127], dtype=torch.int8))
+        result = ShiftIntensity(offset=-1, safe=True)(image)
+        torch.testing.assert_close(result, torch.tensor([-128, -1, 126], dtype=torch.int8))
+
     def test_value(self):
         shifter = ShiftIntensity(offset=1.0)
         result = shifter(self.imt)

From 63f1803787c4e85b96ed57f5f1acde067a67b65c Mon Sep 17 00:00:00 2001
From: "pre-commit-ci[bot]"
 <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Date: Sun, 8 Nov 2026 00:50:50 +0000
Subject: [PATCH 2/2] [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci
---
 tests/transforms/test_shift_intensity.py | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

diff --git a/tests/transforms/test_shift_intensity.py b/tests/transforms/test_shift_intensity.py
index 4b626fc5eb6..4346f9adc90 100644
--- a/tests/transforms/test_shift_intensity.py
+++ b/tests/transforms/test_shift_intensity.py
@@ -23,12 +23,7 @@
 class TestShiftIntensity(NumpyImageTestCase2D):
     def test_safe_integer_offset_clamps_unsigned_image(self):
         image = torch.tensor([0, 100, 255], dtype=torch.uint8)
-        for offset, expected in (
-            (1, [1, 101, 255]),
-            (-1, [0, 99, 254]),
-            (1.0, [1, 101, 255]),
-            (-1.0, [0, 99, 254]),
-        ):
+        for offset, expected in ((1, [1, 101, 255]), (-1, [0, 99, 254]), (1.0, [1, 101, 255]), (-1.0, [0, 99, 254])):
             with self.subTest(offset=offset):
                 result = ShiftIntensity(offset=offset, safe=True)(image)
                 torch.testing.assert_close(result, torch.tensor(expected, dtype=torch.uint8))