3466 3467 Add channel_wise and correct doc-string

monai/transforms/intensity/array.py

@@ -608,8 +608,8 @@ class NormalizeIntensity(Transform):
         subtrahend: the amount to subtract by (usually the mean).
         divisor: the amount to divide by (usually the standard deviation).
         nonzero: whether only normalize non-zero values.
-        channel_wise: if using calculated mean and std, calculate on each channel separately
-            or calculate on the entire image directly.
+        channel_wise: if True, calculate on each channel separately, otherwise, calculate on
+            the entire image directly. default to False.
         dtype: output data type, if None, same as input image. defaults to float32.
     """
 
@@ -919,6 +919,8 @@ class ScaleIntensityRangePercentiles(Transform):
         b_max: intensity target range max.
         clip: whether to perform clip after scaling.
         relative: whether to scale to the corresponding percentiles of [b_min, b_max].
+        channel_wise: if True, compute intensity percentile and normalize every channel separately.
+            default to False.
         dtype: output data type, if None, same as input image. defaults to float32.
     """
 
@@ -932,6 +934,7 @@ def __init__(
         b_max: Optional[float],
         clip: bool = False,
         relative: bool = False,
+        channel_wise: bool = False,
         dtype: DtypeLike = np.float32,
     ) -> None:
         if lower < 0.0 or lower > 100.0:
@@ -944,12 +947,10 @@ def __init__(
         self.b_max = b_max
         self.clip = clip
         self.relative = relative
+        self.channel_wise = channel_wise
         self.dtype = dtype
 
-    def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
-        """
-        Apply the transform to `img`.
-        """
+    def _normalize(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
         a_min: float = percentile(img, self.lower)  # type: ignore
         a_max: float = percentile(img, self.upper)  # type: ignore
         b_min = self.b_min
@@ -967,6 +968,18 @@ def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
         img = scalar(img)
         return img
 
+    def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
+        """
+        Apply the transform to `img`.
+        """
+        if self.channel_wise:
+            for i, d in enumerate(img):
+                img[i] = self._normalize(img=d)  # type: ignore
+        else:
+            img = self._normalize(img=img)
+
+        return img
+
 
 class MaskIntensity(Transform):
     """

monai/transforms/intensity/dictionary.py

@@ -655,8 +655,8 @@ class NormalizeIntensityd(MapTransform):
         subtrahend: the amount to subtract by (usually the mean)
         divisor: the amount to divide by (usually the standard deviation)
         nonzero: whether only normalize non-zero values.
-        channel_wise: if using calculated mean and std, calculate on each channel separately
-            or calculate on the entire image directly.
+        channel_wise: if True, calculate on each channel separately, otherwise, calculate on
+            the entire image directly. default to False.
         dtype: output data type, if None, same as input image. defaults to float32.
         allow_missing_keys: don't raise exception if key is missing.
     """
@@ -844,6 +844,8 @@ class ScaleIntensityRangePercentilesd(MapTransform):
         b_max: intensity target range max.
         clip: whether to perform clip after scaling.
         relative: whether to scale to the corresponding percentiles of [b_min, b_max]
+        channel_wise: if True, compute intensity percentile and normalize every channel separately.
+            default to False.
         dtype: output data type, if None, same as input image. defaults to float32.
         allow_missing_keys: don't raise exception if key is missing.
     """
@@ -859,11 +861,12 @@ def __init__(
         b_max: Optional[float],
         clip: bool = False,
         relative: bool = False,
+        channel_wise: bool = False,
         dtype: DtypeLike = np.float32,
         allow_missing_keys: bool = False,
     ) -> None:
         super().__init__(keys, allow_missing_keys)
-        self.scaler = ScaleIntensityRangePercentiles(lower, upper, b_min, b_max, clip, relative, dtype)
+        self.scaler = ScaleIntensityRangePercentiles(lower, upper, b_min, b_max, clip, relative, channel_wise, dtype)
 
     def __call__(self, data: Mapping[Hashable, NdarrayOrTensor]) -> Dict[Hashable, NdarrayOrTensor]:
         d = dict(data)

monai/transforms/utils_pytorch_numpy_unification.py

@@ -81,16 +81,21 @@ def clip(a: NdarrayOrTensor, a_min, a_max) -> NdarrayOrTensor:
     return result
 
 
-def percentile(x: NdarrayOrTensor, q) -> Union[NdarrayOrTensor, float, int]:
+def percentile(x: NdarrayOrTensor, q, dim: Optional[int] = None) -> Union[NdarrayOrTensor, float, int]:
     """`np.percentile` with equivalent implementation for torch.
 
     Pytorch uses `quantile`, but this functionality is only available from v1.7.
     For earlier methods, we calculate it ourselves. This doesn't do interpolation,
     so is the equivalent of ``numpy.percentile(..., interpolation="nearest")``.
+    For more details, please refer to:
+    https://pytorch.org/docs/stable/generated/torch.quantile.html.
+    https://numpy.org/doc/stable/reference/generated/numpy.percentile.html.
 
     Args:
         x: input data
         q: percentile to compute (should in range 0 <= q <= 100)
+        dim: the dim along which the percentiles are computed. default is to compute the percentile
+            along a flattened version of the array. only work for numpy array or Tensor with PyTorch >= 1.7.0.
 
     Returns:
         Resulting value (scalar)
@@ -102,11 +107,11 @@ def percentile(x: NdarrayOrTensor, q) -> Union[NdarrayOrTensor, float, int]:
         raise ValueError
     result: Union[NdarrayOrTensor, float, int]
     if isinstance(x, np.ndarray):
-        result = np.percentile(x, q)
+        result = np.percentile(x, q, axis=dim)
     else:
         q = torch.tensor(q, device=x.device)
         if hasattr(torch, "quantile"):
-            result = torch.quantile(x, q / 100.0)
+            result = torch.quantile(x, q / 100.0, dim=dim)
         else:
             # Note that ``kthvalue()`` works one-based, i.e., the first sorted value
             # corresponds to k=1, not k=0. Thus, we need the `1 +`.

tests/test_scale_intensity_range_percentiles.py

@@ -19,7 +19,7 @@
 
 class TestScaleIntensityRangePercentiles(NumpyImageTestCase2D):
     def test_scaling(self):
-        img = self.imt
+        img = self.imt[0]
         lower = 10
         upper = 99
         b_min = 0
@@ -34,7 +34,7 @@ def test_scaling(self):
             assert_allclose(result, p(expected), rtol=1e-4)
 
     def test_relative_scaling(self):
-        img = self.imt
+        img = self.imt[0]
         lower = 10
         upper = 99
         b_min = 100
@@ -65,6 +65,26 @@ def test_invalid_instantiation(self):
         self.assertRaises(ValueError, ScaleIntensityRangePercentiles, lower=30, upper=-20, b_min=0, b_max=255)
         self.assertRaises(ValueError, ScaleIntensityRangePercentiles, lower=30, upper=900, b_min=0, b_max=255)
 
+    def test_channel_wise(self):
+        img = self.imt[0]
+        lower = 10
+        upper = 99
+        b_min = 0
+        b_max = 255
+        scaler = ScaleIntensityRangePercentiles(
+            lower=lower, upper=upper, b_min=b_min, b_max=b_max, channel_wise=True, dtype=np.uint8
+        )
+        expected = []
+        for c in img:
+            a_min = np.percentile(c, lower)
+            a_max = np.percentile(c, upper)
+            expected.append(((c - a_min) / (a_max - a_min)) * (b_max - b_min) + b_min)
+        expected = np.stack(expected).astype(np.uint8)
+
+        for p in TEST_NDARRAYS:
+            result = scaler(p(img))
+            assert_allclose(result, p(expected), rtol=1e-4)
+
 
 if __name__ == "__main__":
     unittest.main()

tests/test_scale_intensity_range_percentilesd.py

@@ -27,13 +27,12 @@ def test_scaling(self):
 
         a_min = np.percentile(img, lower)
         a_max = np.percentile(img, upper)
-        expected = (img - a_min) / (a_max - a_min)
-        expected = (expected * (b_max - b_min)) + b_min
+        expected = (((img - a_min) / (a_max - a_min)) * (b_max - b_min) + b_min).astype(np.uint8)
 
         for p in TEST_NDARRAYS:
             data = {"img": p(img)}
             scaler = ScaleIntensityRangePercentilesd(
-                keys=data.keys(), lower=lower, upper=upper, b_min=b_min, b_max=b_max
+                keys=data.keys(), lower=lower, upper=upper, b_min=b_min, b_max=b_max, dtype=np.uint8
             )
             assert_allclose(p(expected), scaler(data)["img"], rtol=1e-4)
 
@@ -75,6 +74,26 @@ def test_invalid_instantiation(self):
             s = ScaleIntensityRangePercentilesd(keys=["img"], lower=30, upper=90, b_min=None, b_max=20, relative=True)
             s(self.imt)
 
+    def test_channel_wise(self):
+        img = self.imt
+        lower = 10
+        upper = 99
+        b_min = 0
+        b_max = 255
+        scaler = ScaleIntensityRangePercentilesd(
+            keys="img", lower=lower, upper=upper, b_min=b_min, b_max=b_max, channel_wise=True, dtype=np.uint8
+        )
+        expected = []
+        for c in img:
+            a_min = np.percentile(c, lower)
+            a_max = np.percentile(c, upper)
+            expected.append((((c - a_min) / (a_max - a_min)) * (b_max - b_min) + b_min).astype(np.uint8))
+        expected = np.stack(expected)
+
+        for p in TEST_NDARRAYS:
+            data = {"img": p(img)}
+            assert_allclose(scaler(data)["img"], p(expected), rtol=1e-4)
+
 
 if __name__ == "__main__":
     unittest.main()

tests/test_utils_pytorch_numpy_unification.py

@@ -16,7 +16,7 @@
 
 from monai.transforms.utils_pytorch_numpy_unification import percentile
 from monai.utils import set_determinism
-from tests.utils import TEST_NDARRAYS, assert_allclose
+from tests.utils import TEST_NDARRAYS, SkipIfBeforePyTorchVersion, assert_allclose
 
 
 class TestPytorchNumpyUnification(unittest.TestCase):
@@ -42,6 +42,18 @@ def test_fails(self):
                 with self.assertRaises(ValueError):
                     percentile(arr, q)
 
+    @SkipIfBeforePyTorchVersion((1, 7))
+    def test_dim(self):
+        q = np.random.randint(0, 100, size=50)
+        results = []
+        for p in TEST_NDARRAYS:
+            arr = p(np.arange(6).reshape(1, 2, 3).astype(np.float32))
+            results.append(percentile(arr, q, dim=1))
+            # pre torch 1.7, no `quantile`. Our own method doesn't interpolate,
+            # so we can only be accurate to 0.5
+            atol = 0.5 if not hasattr(torch, "quantile") else 1e-4
+            assert_allclose(results[0], results[-1], type_test=False, atol=atol)
+
 
 if __name__ == "__main__":
     unittest.main()