From c5f225a94e7f9ca387a7e29fd2252ff6a4d6d10f Mon Sep 17 00:00:00 2001
From: yc7620 <yaniel.cabrera20@imperial.ac.uk>
Date: Thu, 10 Feb 2022 09:36:59 +0200
Subject: [PATCH 1/6] intensity augmentation transforms

Signed-off-by: yc7620 <yaniel.cabrera20@imperial.ac.uk>
---
 monai/config/deviceconfig.py                  |   8 +-
 monai/data/dataset.py                         |   2 +-
 monai/data/dataset_summary.py                 |   2 +-
 monai/handlers/parameter_scheduler.py         |   2 +-
 monai/losses/image_dissimilarity.py           |   6 +-
 monai/networks/blocks/activation.py           |   2 +
 monai/networks/blocks/selfattention.py        |   2 +-
 monai/networks/blocks/upsample.py             |   4 +-
 monai/networks/blocks/warp.py                 |   2 +-
 monai/networks/layers/convutils.py            |   2 +-
 monai/networks/nets/dints.py                  |  10 +-
 monai/networks/nets/highresnet.py             |   2 +-
 monai/networks/nets/regunet.py                |   6 +-
 monai/networks/nets/segresnet.py              |   6 +-
 monai/networks/utils.py                       |   2 +-
 monai/transforms/__init__.py                  |   2 +
 monai/transforms/intensity/array.py           | 128 +++++++++++++++++-
 monai/transforms/smooth_field/array.py        |   2 +-
 .../transforms/utils_create_transform_ims.py  |   2 +-
 tests/test_lmdbdataset.py                     |   6 +-
 tests/test_tile_on_grid.py                    |   2 +-
 tests/test_tile_on_grid_dict.py               |   2 +-
 22 files changed, 164 insertions(+), 38 deletions(-)

diff --git a/monai/config/deviceconfig.py b/monai/config/deviceconfig.py
index fd7ca572e6..91b944bde5 100644
--- a/monai/config/deviceconfig.py
+++ b/monai/config/deviceconfig.py
@@ -161,9 +161,9 @@ def get_system_info() -> OrderedDict:
                 ),
             )
             mem = psutil.virtual_memory()
-            _dict_append(output, "Total physical memory (GB)", lambda: round(mem.total / 1024**3, 1))
-            _dict_append(output, "Available memory (GB)", lambda: round(mem.available / 1024**3, 1))
-            _dict_append(output, "Used memory (GB)", lambda: round(mem.used / 1024**3, 1))
+            _dict_append(output, "Total physical memory (GB)", lambda: round(mem.total / 1024 ** 3, 1))
+            _dict_append(output, "Available memory (GB)", lambda: round(mem.available / 1024 ** 3, 1))
+            _dict_append(output, "Used memory (GB)", lambda: round(mem.used / 1024 ** 3, 1))
 
     return output
 
@@ -209,7 +209,7 @@ def get_gpu_info() -> OrderedDict:
         _dict_append(output, f"GPU {gpu} Is integrated", lambda: bool(gpu_info.is_integrated))
         _dict_append(output, f"GPU {gpu} Is multi GPU board", lambda: bool(gpu_info.is_multi_gpu_board))
         _dict_append(output, f"GPU {gpu} Multi processor count", lambda: gpu_info.multi_processor_count)
-        _dict_append(output, f"GPU {gpu} Total memory (GB)", lambda: round(gpu_info.total_memory / 1024**3, 1))
+        _dict_append(output, f"GPU {gpu} Total memory (GB)", lambda: round(gpu_info.total_memory / 1024 ** 3, 1))
         _dict_append(output, f"GPU {gpu} CUDA capability (maj.min)", lambda: f"{gpu_info.major}.{gpu_info.minor}")
 
     return output
diff --git a/monai/data/dataset.py b/monai/data/dataset.py
index 156708c7dd..8c50446815 100644
--- a/monai/data/dataset.py
+++ b/monai/data/dataset.py
@@ -513,7 +513,7 @@ def __init__(
         self.db_file = self.cache_dir / f"{db_name}.lmdb"
         self.lmdb_kwargs = lmdb_kwargs or {}
         if not self.lmdb_kwargs.get("map_size", 0):
-            self.lmdb_kwargs["map_size"] = 1024**4  # default map_size
+            self.lmdb_kwargs["map_size"] = 1024 ** 4  # default map_size
         # lmdb is single-writer multi-reader by default
         # the cache is created without multi-threading
         self._read_env = None
diff --git a/monai/data/dataset_summary.py b/monai/data/dataset_summary.py
index 956e038569..2b4df4ebbf 100644
--- a/monai/data/dataset_summary.py
+++ b/monai/data/dataset_summary.py
@@ -150,7 +150,7 @@ def calculate_statistics(self, foreground_threshold: int = 0):
 
         self.data_max, self.data_min = max(voxel_max), min(voxel_min)
         self.data_mean = (voxel_sum / voxel_ct).item()
-        self.data_std = (torch.sqrt(voxel_square_sum / voxel_ct - self.data_mean**2)).item()
+        self.data_std = (torch.sqrt(voxel_square_sum / voxel_ct - self.data_mean ** 2)).item()
 
     def calculate_percentiles(
         self,
diff --git a/monai/handlers/parameter_scheduler.py b/monai/handlers/parameter_scheduler.py
index 67c51fd351..c0e18edcd0 100644
--- a/monai/handlers/parameter_scheduler.py
+++ b/monai/handlers/parameter_scheduler.py
@@ -134,7 +134,7 @@ def _exponential(initial_value: float, gamma: float, current_step: int) -> float
         Returns:
             float: new parameter value
         """
-        return initial_value * gamma**current_step
+        return initial_value * gamma ** current_step
 
     @staticmethod
     def _step(initial_value: float, gamma: float, step_size: int, current_step: int) -> float:
diff --git a/monai/losses/image_dissimilarity.py b/monai/losses/image_dissimilarity.py
index a06f6fb5cd..b527522cd7 100644
--- a/monai/losses/image_dissimilarity.py
+++ b/monai/losses/image_dissimilarity.py
@@ -126,7 +126,7 @@ def forward(self, pred: torch.Tensor, target: torch.Tensor) -> torch.Tensor:
         if target.shape != pred.shape:
             raise ValueError(f"ground truth has differing shape ({target.shape}) from pred ({pred.shape})")
 
-        t2, p2, tp = target**2, pred**2, target * pred
+        t2, p2, tp = target ** 2, pred ** 2, target * pred
         kernel, kernel_vol = self.kernel.to(pred), self.kernel_vol.to(pred)
         # sum over kernel
         t_sum = separable_filtering(target, kernels=[kernel.to(pred)] * self.ndim)
@@ -217,7 +217,7 @@ def __init__(
         self.num_bins = num_bins
         self.kernel_type = kernel_type
         if self.kernel_type == "gaussian":
-            self.preterm = 1 / (2 * sigma**2)
+            self.preterm = 1 / (2 * sigma ** 2)
             self.bin_centers = bin_centers[None, None, ...]
         self.smooth_nr = float(smooth_nr)
         self.smooth_dr = float(smooth_dr)
@@ -280,7 +280,7 @@ def parzen_windowing_b_spline(self, img: torch.Tensor, order: int) -> Tuple[torc
             weight = weight + (sample_bin_matrix < 0.5) + (sample_bin_matrix == 0.5) * 0.5
         elif order == 3:
             weight = (
-                weight + (4 - 6 * sample_bin_matrix**2 + 3 * sample_bin_matrix**3) * (sample_bin_matrix < 1) / 6
+                weight + (4 - 6 * sample_bin_matrix ** 2 + 3 * sample_bin_matrix ** 3) * (sample_bin_matrix < 1) / 6
             )
             weight = weight + (2 - sample_bin_matrix) ** 3 * (sample_bin_matrix >= 1) * (sample_bin_matrix < 2) / 6
         else:
diff --git a/monai/networks/blocks/activation.py b/monai/networks/blocks/activation.py
index 1526b37056..cf27b22552 100644
--- a/monai/networks/blocks/activation.py
+++ b/monai/networks/blocks/activation.py
@@ -19,6 +19,7 @@
     def monai_mish(x, inplace: bool = False):
         return torch.nn.functional.mish(x, inplace=inplace)
 
+
 else:
 
     def monai_mish(x, inplace: bool = False):
@@ -30,6 +31,7 @@ def monai_mish(x, inplace: bool = False):
     def monai_swish(x, inplace: bool = False):
         return torch.nn.functional.silu(x, inplace=inplace)
 
+
 else:
 
     def monai_swish(x, inplace: bool = False):
diff --git a/monai/networks/blocks/selfattention.py b/monai/networks/blocks/selfattention.py
index cf837c5a6f..4a86cd84bc 100644
--- a/monai/networks/blocks/selfattention.py
+++ b/monai/networks/blocks/selfattention.py
@@ -46,7 +46,7 @@ def __init__(self, hidden_size: int, num_heads: int, dropout_rate: float = 0.0)
         self.drop_output = nn.Dropout(dropout_rate)
         self.drop_weights = nn.Dropout(dropout_rate)
         self.head_dim = hidden_size // num_heads
-        self.scale = self.head_dim**-0.5
+        self.scale = self.head_dim ** -0.5
 
     def forward(self, x):
         q, k, v = einops.rearrange(self.qkv(x), "b h (qkv l d) -> qkv b l h d", qkv=3, l=self.num_heads)
diff --git a/monai/networks/blocks/upsample.py b/monai/networks/blocks/upsample.py
index fa3929df20..c72d1bc518 100644
--- a/monai/networks/blocks/upsample.py
+++ b/monai/networks/blocks/upsample.py
@@ -219,7 +219,7 @@ def __init__(
             out_channels = out_channels or in_channels
             if not out_channels:
                 raise ValueError("in_channels need to be specified.")
-            conv_out_channels = out_channels * (scale_factor**self.dimensions)
+            conv_out_channels = out_channels * (scale_factor ** self.dimensions)
             self.conv_block = Conv[Conv.CONV, self.dimensions](
                 in_channels=in_channels, out_channels=conv_out_channels, kernel_size=3, stride=1, padding=1, bias=bias
             )
@@ -247,7 +247,7 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
             x: Tensor in shape (batch, channel, spatial_1[, spatial_2, ...).
         """
         x = self.conv_block(x)
-        if x.shape[1] % (self.scale_factor**self.dimensions) != 0:
+        if x.shape[1] % (self.scale_factor ** self.dimensions) != 0:
             raise ValueError(
                 f"Number of channels after `conv_block` ({x.shape[1]}) must be evenly "
                 "divisible by scale_factor ** dimensions "
diff --git a/monai/networks/blocks/warp.py b/monai/networks/blocks/warp.py
index 5b925258b6..9fdaab0a48 100644
--- a/monai/networks/blocks/warp.py
+++ b/monai/networks/blocks/warp.py
@@ -150,7 +150,7 @@ def forward(self, dvf):
         Returns:
             a dense displacement field
         """
-        ddf: torch.Tensor = dvf / (2**self.num_steps)
+        ddf: torch.Tensor = dvf / (2 ** self.num_steps)
         for _ in range(self.num_steps):
             ddf = ddf + self.warp_layer(image=ddf, ddf=ddf)
         return ddf
diff --git a/monai/networks/layers/convutils.py b/monai/networks/layers/convutils.py
index 1e9ce954e8..5efb6e792f 100644
--- a/monai/networks/layers/convutils.py
+++ b/monai/networks/layers/convutils.py
@@ -115,7 +115,7 @@ def gaussian_1d(
         out = out.clamp(min=0)
     elif approx.lower() == "sampled":
         x = torch.arange(-tail, tail + 1, dtype=torch.float, device=sigma.device)
-        out = torch.exp(-0.5 / (sigma * sigma) * x**2)
+        out = torch.exp(-0.5 / (sigma * sigma) * x ** 2)
         if not normalize:  # compute the normalizer
             out = out / (2.5066282 * sigma)
     elif approx.lower() == "scalespace":
diff --git a/monai/networks/nets/dints.py b/monai/networks/nets/dints.py
index a4aaf32eed..c024d6e0f1 100644
--- a/monai/networks/nets/dints.py
+++ b/monai/networks/nets/dints.py
@@ -124,7 +124,7 @@ def __init__(self, in_channel: int, out_channel: int, spatial_dims: int = 3):
         # s0 is upsampled 2x from s1, representing feature sizes at two resolutions.
         # in_channel * s0 (activation) + 3 * out_channel * s1 (convolution, concatenation, normalization)
         # s0 = s1 * 2^(spatial_dims) = output_size / out_channel * 2^(spatial_dims)
-        self.ram_cost = in_channel / out_channel * 2**self._spatial_dims + 3
+        self.ram_cost = in_channel / out_channel * 2 ** self._spatial_dims + 3
 
 
 class MixedOp(nn.Module):
@@ -330,7 +330,7 @@ def __init__(
             # define downsample stems before DiNTS search
             if use_downsample:
                 self.stem_down[str(res_idx)] = StemTS(
-                    nn.Upsample(scale_factor=1 / (2**res_idx), mode=mode, align_corners=True),
+                    nn.Upsample(scale_factor=1 / (2 ** res_idx), mode=mode, align_corners=True),
                     conv_type(
                         in_channels=in_channels,
                         out_channels=self.filter_nums[res_idx],
@@ -373,7 +373,7 @@ def __init__(
 
             else:
                 self.stem_down[str(res_idx)] = StemTS(
-                    nn.Upsample(scale_factor=1 / (2**res_idx), mode=mode, align_corners=True),
+                    nn.Upsample(scale_factor=1 / (2 ** res_idx), mode=mode, align_corners=True),
                     conv_type(
                         in_channels=in_channels,
                         out_channels=self.filter_nums[res_idx],
@@ -789,7 +789,7 @@ def get_ram_cost_usage(self, in_size, full: bool = False):
         image_size = np.array(in_size[-self._spatial_dims :])
         sizes = []
         for res_idx in range(self.num_depths):
-            sizes.append(batch_size * self.filter_nums[res_idx] * (image_size // (2**res_idx)).prod())
+            sizes.append(batch_size * self.filter_nums[res_idx] * (image_size // (2 ** res_idx)).prod())
         sizes = torch.tensor(sizes).to(torch.float32).to(self.device) / (2 ** (int(self.use_downsample)))
         probs_a, arch_code_prob_a = self.get_prob_a(child=False)
         cell_prob = F.softmax(self.log_alpha_c, dim=-1)
@@ -807,7 +807,7 @@ def get_ram_cost_usage(self, in_size, full: bool = False):
                     * (1 + (ram_cost[blk_idx, path_idx] * cell_prob[blk_idx, path_idx]).sum())
                     * sizes[self.arch_code2out[path_idx]]
                 )
-        return usage * 32 / 8 / 1024**2
+        return usage * 32 / 8 / 1024 ** 2
 
     def get_topology_entropy(self, probs):
         """
diff --git a/monai/networks/nets/highresnet.py b/monai/networks/nets/highresnet.py
index 891a65e67b..95c0c758af 100644
--- a/monai/networks/nets/highresnet.py
+++ b/monai/networks/nets/highresnet.py
@@ -168,7 +168,7 @@ def __init__(
         # residual blocks
         for (idx, params) in enumerate(layer_params[1:-2]):  # res blocks except the 1st and last two conv layers.
             _in_chns, _out_chns = _out_chns, params["n_features"]
-            _dilation = 2**idx
+            _dilation = 2 ** idx
             for _ in range(params["repeat"]):
                 blocks.append(
                     HighResBlock(
diff --git a/monai/networks/nets/regunet.py b/monai/networks/nets/regunet.py
index 6776c7ce9e..8524563faa 100644
--- a/monai/networks/nets/regunet.py
+++ b/monai/networks/nets/regunet.py
@@ -92,7 +92,7 @@ def __init__(
             raise AssertionError
         self.encode_kernel_sizes: List[int] = encode_kernel_sizes
 
-        self.num_channels = [self.num_channel_initial * (2**d) for d in range(self.depth + 1)]
+        self.num_channels = [self.num_channel_initial * (2 ** d) for d in range(self.depth + 1)]
         self.min_extract_level = min(self.extract_levels)
 
         # init layers
@@ -310,14 +310,14 @@ def __init__(
         encode_kernel_sizes: Union[int, List[int]] = 3,
     ):
         for size in image_size:
-            if size % (2**depth) != 0:
+            if size % (2 ** depth) != 0:
                 raise ValueError(
                     f"given depth {depth}, "
                     f"all input spatial dimension must be divisible by {2 ** depth}, "
                     f"got input of size {image_size}"
                 )
         self.image_size = image_size
-        self.decode_size = [size // (2**depth) for size in image_size]
+        self.decode_size = [size // (2 ** depth) for size in image_size]
         super().__init__(
             spatial_dims=spatial_dims,
             in_channels=in_channels,
diff --git a/monai/networks/nets/segresnet.py b/monai/networks/nets/segresnet.py
index 299f1ca811..d2c45dd3a3 100644
--- a/monai/networks/nets/segresnet.py
+++ b/monai/networks/nets/segresnet.py
@@ -102,7 +102,7 @@ def _make_down_layers(self):
         down_layers = nn.ModuleList()
         blocks_down, spatial_dims, filters, norm = (self.blocks_down, self.spatial_dims, self.init_filters, self.norm)
         for i in range(len(blocks_down)):
-            layer_in_channels = filters * 2**i
+            layer_in_channels = filters * 2 ** i
             pre_conv = (
                 get_conv_layer(spatial_dims, layer_in_channels // 2, layer_in_channels, stride=2)
                 if i > 0
@@ -299,12 +299,12 @@ def _get_vae_loss(self, net_input: torch.Tensor, vae_input: torch.Tensor):
         if self.vae_estimate_std:
             z_sigma = self.vae_fc2(x_vae)
             z_sigma = F.softplus(z_sigma)
-            vae_reg_loss = 0.5 * torch.mean(z_mean**2 + z_sigma**2 - torch.log(1e-8 + z_sigma**2) - 1)
+            vae_reg_loss = 0.5 * torch.mean(z_mean ** 2 + z_sigma ** 2 - torch.log(1e-8 + z_sigma ** 2) - 1)
 
             x_vae = z_mean + z_sigma * z_mean_rand
         else:
             z_sigma = self.vae_default_std
-            vae_reg_loss = torch.mean(z_mean**2)
+            vae_reg_loss = torch.mean(z_mean ** 2)
 
             x_vae = z_mean + z_sigma * z_mean_rand
 
diff --git a/monai/networks/utils.py b/monai/networks/utils.py
index 47cc838fdc..a4ca0a6fd5 100644
--- a/monai/networks/utils.py
+++ b/monai/networks/utils.py
@@ -270,7 +270,7 @@ def pixelshuffle(
     dim, factor = spatial_dims, scale_factor
     input_size = list(x.size())
     batch_size, channels = input_size[:2]
-    scale_divisor = factor**dim
+    scale_divisor = factor ** dim
 
     if channels % scale_divisor != 0:
         raise ValueError(
diff --git a/monai/transforms/__init__.py b/monai/transforms/__init__.py
index 26601de76b..b90140b455 100644
--- a/monai/transforms/__init__.py
+++ b/monai/transforms/__init__.py
@@ -85,6 +85,7 @@
     GaussianSmooth,
     GibbsNoise,
     HistogramNormalize,
+    IntensityRemap,
     KSpaceSpikeNoise,
     MaskIntensity,
     NormalizeIntensity,
@@ -98,6 +99,7 @@
     RandGaussianSmooth,
     RandGibbsNoise,
     RandHistogramShift,
+    RandIntensityRemap,
     RandKSpaceSpikeNoise,
     RandRicianNoise,
     RandScaleIntensity,
diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index b5cf61ce0c..7a50a528d0 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -16,7 +16,7 @@
 from abc import abstractmethod
 from collections.abc import Iterable
 from functools import partial
-from typing import Any, Callable, List, Optional, Sequence, Tuple, Union
+from typing import Any, Callable, List, Mapping, Optional, Sequence, Tuple, Union
 from warnings import warn
 
 import numpy as np
@@ -75,6 +75,8 @@
     "RandCoarseDropout",
     "RandCoarseShuffle",
     "HistogramNormalize",
+    "IntensityRemap",
+    "RandIntensityRemap",
 ]
 
 
@@ -182,9 +184,9 @@ def _add_noise(self, img: NdarrayOrTensor, mean: float, std: float):
         if isinstance(img, torch.Tensor):
             n1 = torch.tensor(self._noise1, device=img.device)
             n2 = torch.tensor(self._noise2, device=img.device)
-            return torch.sqrt((img + n1) ** 2 + n2**2)
+            return torch.sqrt((img + n1) ** 2 + n2 ** 2)
 
-        return np.sqrt((img + self._noise1) ** 2 + self._noise2**2)
+        return np.sqrt((img + self._noise1) ** 2 + self._noise2 ** 2)
 
     def __call__(self, img: NdarrayOrTensor, randomize: bool = True) -> NdarrayOrTensor:
         """
@@ -2053,3 +2055,123 @@ def __call__(self, img: NdarrayOrTensor, mask: Optional[NdarrayOrTensor] = None)
         out, *_ = convert_to_dst_type(src=ret, dst=img, dtype=self.dtype or img.dtype)
 
         return out
+
+
+class IntensityRemap(RandomizableTransform):
+    """
+    Transform for intensity remapping of images. The intensity at each
+    pixel is replaced by a new values coming from an intensity remappping
+    curve.
+
+    The remapping curve is created by uniformly sampling values from the
+    possible intensities for the input image and then adding a linear
+    component. The curve is the rescaled to the input image intensity range.
+
+    Intended to be used as a means to data augmentation via:
+    :py:class:`monai.transforms.RandIntensityRemap`.
+
+    Implementation is described in the work:
+    `Intensity augmentation for domain transfer of whole breast segmentation
+    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
+
+    Note: for large images the transform may be too slow to apply on the fly. 
+
+    Args:
+        kernel_size: window size for averaging operation for the remapping
+            curve.
+        slope: slope of the linear component. Easiest to leave default value
+            and tune the kernel_size parameter instead.
+        return_map: set to True for the transform to return a dictionary version
+            of the lookup table used in the intensity remapping. The keys
+            correspond to the old intensities, and the values are the new
+            values.
+    """
+
+    def __init__(self, kernel_size: int = 30, slope: float = 0.7, return_map: bool = False):
+
+        super().__init__()
+
+        self.kernel_size = kernel_size
+        self.slope = slope
+        self.return_map = return_map
+
+    def __call__(self, img: torch.Tensor) -> Union[torch.Tensor, Tuple[torch.Tensor, Mapping]]:
+        """
+        Args:
+            img: image to remap.
+        """
+
+        img = img.clone()
+        # sample noise
+        vals_to_sample = torch.unique(img).tolist()
+        noise = torch.from_numpy(self.R.choice(vals_to_sample, len(vals_to_sample) - 1 + self.kernel_size))
+        # smooth
+        noise = torch.nn.AvgPool1d(self.kernel_size, stride=1)(noise.unsqueeze(0)).squeeze()
+        # add linear component
+        grid = torch.arange(len(noise)) / len(noise)
+        noise += self.slope * grid
+        # rescale
+        noise = (noise - noise.min()) / (noise.max() - noise.min()) * img.max() + img.min()
+
+        # intensity remapping function
+        map = dict(zip(vals_to_sample, noise.tolist()))
+
+        for key in map:
+            img[img == key] = map[key]
+
+        if self.return_map:
+            return img, map
+        else:
+            return img
+
+
+class RandIntensityRemap(RandomizableTransform):
+    """
+    Transform for intensity remapping of images. The intensity at each
+    pixel is replaced by a new values coming from an intensity remappping
+    curve.
+
+    The remapping curve is created by uniformly sampling values from the
+    possible intensities for the input image and then adding a linear
+    component. The curve is the rescaled to the input image intensity range.
+
+    Implementation is described in the work:
+    `Intensity augmentation for domain transfer of whole breast segmentation
+    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
+
+    Note: for large images the transform may be too slow to apply on the fly.
+    
+    Args:
+        prob: probability of applying the transform.
+        kernel_size: window size for averaging operation for the remapping
+            curve.
+        slope: slope of the linear component. Easiest to leave default value
+            and tune the kernel_size parameter instead.
+        channel_wise: set to True to treat each channel independently.
+    """
+
+    def __init__(self, prob: float = 0.1, kernel_size: int = 30, slope: float = 0.7, channel_wise: bool = True):
+
+        RandomizableTransform.__init__(self, prob=prob)
+        self.kernel_size = kernel_size
+        self.slope = slope
+        self.channel_wise = True
+
+    def __call__(self, img: torch.Tensor) -> torch.Tensor:
+        """
+        Args:
+            img: image to remap.
+        """
+        super().randomize(None)
+        if self._do_transform:
+            if self.channel_wise:
+                img = torch.stack(
+                    [
+                        IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i])
+                        for i in range(len(img))
+                    ]
+                )
+            else:
+                img = IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img)
+
+        return img
diff --git a/monai/transforms/smooth_field/array.py b/monai/transforms/smooth_field/array.py
index f581687ea5..61342b8941 100644
--- a/monai/transforms/smooth_field/array.py
+++ b/monai/transforms/smooth_field/array.py
@@ -232,7 +232,7 @@ def __call__(self, img: NdarrayOrTensor, randomize: bool = True) -> NdarrayOrTen
         # everything below here is to be computed using the destination type (numpy, tensor, etc.)
 
         img = (img - img_min) / (img_rng + 1e-10)  # rescale to unit values
-        img = img**rfield  # contrast is changed by raising image data to a power, in this case the field
+        img = img ** rfield  # contrast is changed by raising image data to a power, in this case the field
 
         out = (img * img_rng) + img_min  # rescale back to the original image value range
 
diff --git a/monai/transforms/utils_create_transform_ims.py b/monai/transforms/utils_create_transform_ims.py
index b096e1b93d..9fc9206fac 100644
--- a/monai/transforms/utils_create_transform_ims.py
+++ b/monai/transforms/utils_create_transform_ims.py
@@ -383,7 +383,7 @@ def get_images(data, is_label=False):
         # we might need to panel the images. this happens if a transform produces e.g. 4 output images.
         # In this case, we create a 2-by-2 grid from them. Output will be a list containing n_orthog_views,
         # each element being either the image (if num_samples is 1) or the panelled image.
-        nrows = int(np.floor(num_samples**0.5))
+        nrows = int(np.floor(num_samples ** 0.5))
         for view in range(num_orthog_views):
             result = np.asarray([d[view] for d in data])
             nindex, height, width = result.shape
diff --git a/tests/test_lmdbdataset.py b/tests/test_lmdbdataset.py
index 33f27ee4bc..b624e5c4e3 100644
--- a/tests/test_lmdbdataset.py
+++ b/tests/test_lmdbdataset.py
@@ -57,7 +57,7 @@
         SimulateDelayd(keys=["image", "label", "extra"], delay_time=[1e-7, 1e-6, 1e-5]),
     ],
     (128, 128, 128),
-    {"pickle_protocol": 2, "lmdb_kwargs": {"map_size": 100 * 1024**2}},
+    {"pickle_protocol": 2, "lmdb_kwargs": {"map_size": 100 * 1024 ** 2}},
 ]
 
 TEST_CASE_6 = [
@@ -66,7 +66,7 @@
         SimulateDelayd(keys=["image", "label", "extra"], delay_time=[1e-7, 1e-6, 1e-5]),
     ],
     (128, 128, 128),
-    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 100 * 1024**2}},
+    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 100 * 1024 ** 2}},
 ]
 
 TEST_CASE_7 = [
@@ -75,7 +75,7 @@
         SimulateDelayd(keys=["image", "label", "extra"], delay_time=[1e-7, 1e-6, 1e-5]),
     ],
     (128, 128, 128),
-    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 2 * 1024**2}},
+    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 2 * 1024 ** 2}},
 ]
 
 
diff --git a/tests/test_tile_on_grid.py b/tests/test_tile_on_grid.py
index 09434de5e0..212f461c9a 100644
--- a/tests/test_tile_on_grid.py
+++ b/tests/test_tile_on_grid.py
@@ -107,7 +107,7 @@ def make_image(
 
     tiles = np.stack(tiles_list, axis=0)
 
-    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count**2:
+    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count ** 2:
         tiles = tiles[np.argsort(tiles.sum(axis=(1, 2, 3)))]
 
     return imlarge, tiles
diff --git a/tests/test_tile_on_grid_dict.py b/tests/test_tile_on_grid_dict.py
index c6f35fe738..609601f222 100644
--- a/tests/test_tile_on_grid_dict.py
+++ b/tests/test_tile_on_grid_dict.py
@@ -116,7 +116,7 @@ def make_image(
 
     tiles = np.stack(tiles_list, axis=0)
 
-    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count**2:
+    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count ** 2:
         tiles = tiles[np.argsort(tiles.sum(axis=(1, 2, 3)))]
 
     return imlarge, tiles

From 92043636475c5e3d14f710a32a02ced42a0a72d4 Mon Sep 17 00:00:00 2001
From: "pre-commit-ci[bot]"
 <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Date: Thu, 10 Feb 2022 07:58:33 +0000
Subject: [PATCH 2/6] [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci
---
 monai/transforms/intensity/array.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index 7a50a528d0..45307ab918 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -2074,7 +2074,7 @@ class IntensityRemap(RandomizableTransform):
     `Intensity augmentation for domain transfer of whole breast segmentation
     in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
 
-    Note: for large images the transform may be too slow to apply on the fly. 
+    Note: for large images the transform may be too slow to apply on the fly.
 
     Args:
         kernel_size: window size for averaging operation for the remapping
@@ -2140,7 +2140,7 @@ class RandIntensityRemap(RandomizableTransform):
     in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
 
     Note: for large images the transform may be too slow to apply on the fly.
-    
+
     Args:
         prob: probability of applying the transform.
         kernel_size: window size for averaging operation for the remapping

From 26122560b0a67429015e7734fbfd7731dd537c43 Mon Sep 17 00:00:00 2001
From: monai-bot <monai.miccai2019@gmail.com>
Date: Thu, 10 Feb 2022 08:23:06 +0000
Subject: [PATCH 3/6] [MONAI] python code formatting

Signed-off-by: monai-bot <monai.miccai2019@gmail.com>
---
 monai/config/deviceconfig.py                   |  8 ++++----
 monai/data/dataset.py                          |  2 +-
 monai/data/dataset_summary.py                  |  2 +-
 monai/handlers/parameter_scheduler.py          |  2 +-
 monai/losses/image_dissimilarity.py            |  6 +++---
 monai/networks/blocks/activation.py            |  2 --
 monai/networks/blocks/selfattention.py         |  2 +-
 monai/networks/blocks/upsample.py              |  4 ++--
 monai/networks/blocks/warp.py                  |  2 +-
 monai/networks/layers/convutils.py             |  2 +-
 monai/networks/nets/dints.py                   | 10 +++++-----
 monai/networks/nets/highresnet.py              |  2 +-
 monai/networks/nets/regunet.py                 |  6 +++---
 monai/networks/nets/segresnet.py               |  6 +++---
 monai/networks/utils.py                        |  2 +-
 monai/transforms/intensity/array.py            |  4 ++--
 monai/transforms/smooth_field/array.py         |  2 +-
 monai/transforms/utils_create_transform_ims.py |  2 +-
 tests/test_lmdbdataset.py                      |  6 +++---
 tests/test_tile_on_grid.py                     |  2 +-
 tests/test_tile_on_grid_dict.py                |  2 +-
 21 files changed, 37 insertions(+), 39 deletions(-)

diff --git a/monai/config/deviceconfig.py b/monai/config/deviceconfig.py
index 91b944bde5..fd7ca572e6 100644
--- a/monai/config/deviceconfig.py
+++ b/monai/config/deviceconfig.py
@@ -161,9 +161,9 @@ def get_system_info() -> OrderedDict:
                 ),
             )
             mem = psutil.virtual_memory()
-            _dict_append(output, "Total physical memory (GB)", lambda: round(mem.total / 1024 ** 3, 1))
-            _dict_append(output, "Available memory (GB)", lambda: round(mem.available / 1024 ** 3, 1))
-            _dict_append(output, "Used memory (GB)", lambda: round(mem.used / 1024 ** 3, 1))
+            _dict_append(output, "Total physical memory (GB)", lambda: round(mem.total / 1024**3, 1))
+            _dict_append(output, "Available memory (GB)", lambda: round(mem.available / 1024**3, 1))
+            _dict_append(output, "Used memory (GB)", lambda: round(mem.used / 1024**3, 1))
 
     return output
 
@@ -209,7 +209,7 @@ def get_gpu_info() -> OrderedDict:
         _dict_append(output, f"GPU {gpu} Is integrated", lambda: bool(gpu_info.is_integrated))
         _dict_append(output, f"GPU {gpu} Is multi GPU board", lambda: bool(gpu_info.is_multi_gpu_board))
         _dict_append(output, f"GPU {gpu} Multi processor count", lambda: gpu_info.multi_processor_count)
-        _dict_append(output, f"GPU {gpu} Total memory (GB)", lambda: round(gpu_info.total_memory / 1024 ** 3, 1))
+        _dict_append(output, f"GPU {gpu} Total memory (GB)", lambda: round(gpu_info.total_memory / 1024**3, 1))
         _dict_append(output, f"GPU {gpu} CUDA capability (maj.min)", lambda: f"{gpu_info.major}.{gpu_info.minor}")
 
     return output
diff --git a/monai/data/dataset.py b/monai/data/dataset.py
index 8c50446815..156708c7dd 100644
--- a/monai/data/dataset.py
+++ b/monai/data/dataset.py
@@ -513,7 +513,7 @@ def __init__(
         self.db_file = self.cache_dir / f"{db_name}.lmdb"
         self.lmdb_kwargs = lmdb_kwargs or {}
         if not self.lmdb_kwargs.get("map_size", 0):
-            self.lmdb_kwargs["map_size"] = 1024 ** 4  # default map_size
+            self.lmdb_kwargs["map_size"] = 1024**4  # default map_size
         # lmdb is single-writer multi-reader by default
         # the cache is created without multi-threading
         self._read_env = None
diff --git a/monai/data/dataset_summary.py b/monai/data/dataset_summary.py
index 2b4df4ebbf..956e038569 100644
--- a/monai/data/dataset_summary.py
+++ b/monai/data/dataset_summary.py
@@ -150,7 +150,7 @@ def calculate_statistics(self, foreground_threshold: int = 0):
 
         self.data_max, self.data_min = max(voxel_max), min(voxel_min)
         self.data_mean = (voxel_sum / voxel_ct).item()
-        self.data_std = (torch.sqrt(voxel_square_sum / voxel_ct - self.data_mean ** 2)).item()
+        self.data_std = (torch.sqrt(voxel_square_sum / voxel_ct - self.data_mean**2)).item()
 
     def calculate_percentiles(
         self,
diff --git a/monai/handlers/parameter_scheduler.py b/monai/handlers/parameter_scheduler.py
index c0e18edcd0..67c51fd351 100644
--- a/monai/handlers/parameter_scheduler.py
+++ b/monai/handlers/parameter_scheduler.py
@@ -134,7 +134,7 @@ def _exponential(initial_value: float, gamma: float, current_step: int) -> float
         Returns:
             float: new parameter value
         """
-        return initial_value * gamma ** current_step
+        return initial_value * gamma**current_step
 
     @staticmethod
     def _step(initial_value: float, gamma: float, step_size: int, current_step: int) -> float:
diff --git a/monai/losses/image_dissimilarity.py b/monai/losses/image_dissimilarity.py
index b527522cd7..a06f6fb5cd 100644
--- a/monai/losses/image_dissimilarity.py
+++ b/monai/losses/image_dissimilarity.py
@@ -126,7 +126,7 @@ def forward(self, pred: torch.Tensor, target: torch.Tensor) -> torch.Tensor:
         if target.shape != pred.shape:
             raise ValueError(f"ground truth has differing shape ({target.shape}) from pred ({pred.shape})")
 
-        t2, p2, tp = target ** 2, pred ** 2, target * pred
+        t2, p2, tp = target**2, pred**2, target * pred
         kernel, kernel_vol = self.kernel.to(pred), self.kernel_vol.to(pred)
         # sum over kernel
         t_sum = separable_filtering(target, kernels=[kernel.to(pred)] * self.ndim)
@@ -217,7 +217,7 @@ def __init__(
         self.num_bins = num_bins
         self.kernel_type = kernel_type
         if self.kernel_type == "gaussian":
-            self.preterm = 1 / (2 * sigma ** 2)
+            self.preterm = 1 / (2 * sigma**2)
             self.bin_centers = bin_centers[None, None, ...]
         self.smooth_nr = float(smooth_nr)
         self.smooth_dr = float(smooth_dr)
@@ -280,7 +280,7 @@ def parzen_windowing_b_spline(self, img: torch.Tensor, order: int) -> Tuple[torc
             weight = weight + (sample_bin_matrix < 0.5) + (sample_bin_matrix == 0.5) * 0.5
         elif order == 3:
             weight = (
-                weight + (4 - 6 * sample_bin_matrix ** 2 + 3 * sample_bin_matrix ** 3) * (sample_bin_matrix < 1) / 6
+                weight + (4 - 6 * sample_bin_matrix**2 + 3 * sample_bin_matrix**3) * (sample_bin_matrix < 1) / 6
             )
             weight = weight + (2 - sample_bin_matrix) ** 3 * (sample_bin_matrix >= 1) * (sample_bin_matrix < 2) / 6
         else:
diff --git a/monai/networks/blocks/activation.py b/monai/networks/blocks/activation.py
index cf27b22552..1526b37056 100644
--- a/monai/networks/blocks/activation.py
+++ b/monai/networks/blocks/activation.py
@@ -19,7 +19,6 @@
     def monai_mish(x, inplace: bool = False):
         return torch.nn.functional.mish(x, inplace=inplace)
 
-
 else:
 
     def monai_mish(x, inplace: bool = False):
@@ -31,7 +30,6 @@ def monai_mish(x, inplace: bool = False):
     def monai_swish(x, inplace: bool = False):
         return torch.nn.functional.silu(x, inplace=inplace)
 
-
 else:
 
     def monai_swish(x, inplace: bool = False):
diff --git a/monai/networks/blocks/selfattention.py b/monai/networks/blocks/selfattention.py
index 4a86cd84bc..cf837c5a6f 100644
--- a/monai/networks/blocks/selfattention.py
+++ b/monai/networks/blocks/selfattention.py
@@ -46,7 +46,7 @@ def __init__(self, hidden_size: int, num_heads: int, dropout_rate: float = 0.0)
         self.drop_output = nn.Dropout(dropout_rate)
         self.drop_weights = nn.Dropout(dropout_rate)
         self.head_dim = hidden_size // num_heads
-        self.scale = self.head_dim ** -0.5
+        self.scale = self.head_dim**-0.5
 
     def forward(self, x):
         q, k, v = einops.rearrange(self.qkv(x), "b h (qkv l d) -> qkv b l h d", qkv=3, l=self.num_heads)
diff --git a/monai/networks/blocks/upsample.py b/monai/networks/blocks/upsample.py
index c72d1bc518..fa3929df20 100644
--- a/monai/networks/blocks/upsample.py
+++ b/monai/networks/blocks/upsample.py
@@ -219,7 +219,7 @@ def __init__(
             out_channels = out_channels or in_channels
             if not out_channels:
                 raise ValueError("in_channels need to be specified.")
-            conv_out_channels = out_channels * (scale_factor ** self.dimensions)
+            conv_out_channels = out_channels * (scale_factor**self.dimensions)
             self.conv_block = Conv[Conv.CONV, self.dimensions](
                 in_channels=in_channels, out_channels=conv_out_channels, kernel_size=3, stride=1, padding=1, bias=bias
             )
@@ -247,7 +247,7 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
             x: Tensor in shape (batch, channel, spatial_1[, spatial_2, ...).
         """
         x = self.conv_block(x)
-        if x.shape[1] % (self.scale_factor ** self.dimensions) != 0:
+        if x.shape[1] % (self.scale_factor**self.dimensions) != 0:
             raise ValueError(
                 f"Number of channels after `conv_block` ({x.shape[1]}) must be evenly "
                 "divisible by scale_factor ** dimensions "
diff --git a/monai/networks/blocks/warp.py b/monai/networks/blocks/warp.py
index 9fdaab0a48..5b925258b6 100644
--- a/monai/networks/blocks/warp.py
+++ b/monai/networks/blocks/warp.py
@@ -150,7 +150,7 @@ def forward(self, dvf):
         Returns:
             a dense displacement field
         """
-        ddf: torch.Tensor = dvf / (2 ** self.num_steps)
+        ddf: torch.Tensor = dvf / (2**self.num_steps)
         for _ in range(self.num_steps):
             ddf = ddf + self.warp_layer(image=ddf, ddf=ddf)
         return ddf
diff --git a/monai/networks/layers/convutils.py b/monai/networks/layers/convutils.py
index 5efb6e792f..1e9ce954e8 100644
--- a/monai/networks/layers/convutils.py
+++ b/monai/networks/layers/convutils.py
@@ -115,7 +115,7 @@ def gaussian_1d(
         out = out.clamp(min=0)
     elif approx.lower() == "sampled":
         x = torch.arange(-tail, tail + 1, dtype=torch.float, device=sigma.device)
-        out = torch.exp(-0.5 / (sigma * sigma) * x ** 2)
+        out = torch.exp(-0.5 / (sigma * sigma) * x**2)
         if not normalize:  # compute the normalizer
             out = out / (2.5066282 * sigma)
     elif approx.lower() == "scalespace":
diff --git a/monai/networks/nets/dints.py b/monai/networks/nets/dints.py
index c024d6e0f1..a4aaf32eed 100644
--- a/monai/networks/nets/dints.py
+++ b/monai/networks/nets/dints.py
@@ -124,7 +124,7 @@ def __init__(self, in_channel: int, out_channel: int, spatial_dims: int = 3):
         # s0 is upsampled 2x from s1, representing feature sizes at two resolutions.
         # in_channel * s0 (activation) + 3 * out_channel * s1 (convolution, concatenation, normalization)
         # s0 = s1 * 2^(spatial_dims) = output_size / out_channel * 2^(spatial_dims)
-        self.ram_cost = in_channel / out_channel * 2 ** self._spatial_dims + 3
+        self.ram_cost = in_channel / out_channel * 2**self._spatial_dims + 3
 
 
 class MixedOp(nn.Module):
@@ -330,7 +330,7 @@ def __init__(
             # define downsample stems before DiNTS search
             if use_downsample:
                 self.stem_down[str(res_idx)] = StemTS(
-                    nn.Upsample(scale_factor=1 / (2 ** res_idx), mode=mode, align_corners=True),
+                    nn.Upsample(scale_factor=1 / (2**res_idx), mode=mode, align_corners=True),
                     conv_type(
                         in_channels=in_channels,
                         out_channels=self.filter_nums[res_idx],
@@ -373,7 +373,7 @@ def __init__(
 
             else:
                 self.stem_down[str(res_idx)] = StemTS(
-                    nn.Upsample(scale_factor=1 / (2 ** res_idx), mode=mode, align_corners=True),
+                    nn.Upsample(scale_factor=1 / (2**res_idx), mode=mode, align_corners=True),
                     conv_type(
                         in_channels=in_channels,
                         out_channels=self.filter_nums[res_idx],
@@ -789,7 +789,7 @@ def get_ram_cost_usage(self, in_size, full: bool = False):
         image_size = np.array(in_size[-self._spatial_dims :])
         sizes = []
         for res_idx in range(self.num_depths):
-            sizes.append(batch_size * self.filter_nums[res_idx] * (image_size // (2 ** res_idx)).prod())
+            sizes.append(batch_size * self.filter_nums[res_idx] * (image_size // (2**res_idx)).prod())
         sizes = torch.tensor(sizes).to(torch.float32).to(self.device) / (2 ** (int(self.use_downsample)))
         probs_a, arch_code_prob_a = self.get_prob_a(child=False)
         cell_prob = F.softmax(self.log_alpha_c, dim=-1)
@@ -807,7 +807,7 @@ def get_ram_cost_usage(self, in_size, full: bool = False):
                     * (1 + (ram_cost[blk_idx, path_idx] * cell_prob[blk_idx, path_idx]).sum())
                     * sizes[self.arch_code2out[path_idx]]
                 )
-        return usage * 32 / 8 / 1024 ** 2
+        return usage * 32 / 8 / 1024**2
 
     def get_topology_entropy(self, probs):
         """
diff --git a/monai/networks/nets/highresnet.py b/monai/networks/nets/highresnet.py
index 95c0c758af..891a65e67b 100644
--- a/monai/networks/nets/highresnet.py
+++ b/monai/networks/nets/highresnet.py
@@ -168,7 +168,7 @@ def __init__(
         # residual blocks
         for (idx, params) in enumerate(layer_params[1:-2]):  # res blocks except the 1st and last two conv layers.
             _in_chns, _out_chns = _out_chns, params["n_features"]
-            _dilation = 2 ** idx
+            _dilation = 2**idx
             for _ in range(params["repeat"]):
                 blocks.append(
                     HighResBlock(
diff --git a/monai/networks/nets/regunet.py b/monai/networks/nets/regunet.py
index 8524563faa..6776c7ce9e 100644
--- a/monai/networks/nets/regunet.py
+++ b/monai/networks/nets/regunet.py
@@ -92,7 +92,7 @@ def __init__(
             raise AssertionError
         self.encode_kernel_sizes: List[int] = encode_kernel_sizes
 
-        self.num_channels = [self.num_channel_initial * (2 ** d) for d in range(self.depth + 1)]
+        self.num_channels = [self.num_channel_initial * (2**d) for d in range(self.depth + 1)]
         self.min_extract_level = min(self.extract_levels)
 
         # init layers
@@ -310,14 +310,14 @@ def __init__(
         encode_kernel_sizes: Union[int, List[int]] = 3,
     ):
         for size in image_size:
-            if size % (2 ** depth) != 0:
+            if size % (2**depth) != 0:
                 raise ValueError(
                     f"given depth {depth}, "
                     f"all input spatial dimension must be divisible by {2 ** depth}, "
                     f"got input of size {image_size}"
                 )
         self.image_size = image_size
-        self.decode_size = [size // (2 ** depth) for size in image_size]
+        self.decode_size = [size // (2**depth) for size in image_size]
         super().__init__(
             spatial_dims=spatial_dims,
             in_channels=in_channels,
diff --git a/monai/networks/nets/segresnet.py b/monai/networks/nets/segresnet.py
index d2c45dd3a3..299f1ca811 100644
--- a/monai/networks/nets/segresnet.py
+++ b/monai/networks/nets/segresnet.py
@@ -102,7 +102,7 @@ def _make_down_layers(self):
         down_layers = nn.ModuleList()
         blocks_down, spatial_dims, filters, norm = (self.blocks_down, self.spatial_dims, self.init_filters, self.norm)
         for i in range(len(blocks_down)):
-            layer_in_channels = filters * 2 ** i
+            layer_in_channels = filters * 2**i
             pre_conv = (
                 get_conv_layer(spatial_dims, layer_in_channels // 2, layer_in_channels, stride=2)
                 if i > 0
@@ -299,12 +299,12 @@ def _get_vae_loss(self, net_input: torch.Tensor, vae_input: torch.Tensor):
         if self.vae_estimate_std:
             z_sigma = self.vae_fc2(x_vae)
             z_sigma = F.softplus(z_sigma)
-            vae_reg_loss = 0.5 * torch.mean(z_mean ** 2 + z_sigma ** 2 - torch.log(1e-8 + z_sigma ** 2) - 1)
+            vae_reg_loss = 0.5 * torch.mean(z_mean**2 + z_sigma**2 - torch.log(1e-8 + z_sigma**2) - 1)
 
             x_vae = z_mean + z_sigma * z_mean_rand
         else:
             z_sigma = self.vae_default_std
-            vae_reg_loss = torch.mean(z_mean ** 2)
+            vae_reg_loss = torch.mean(z_mean**2)
 
             x_vae = z_mean + z_sigma * z_mean_rand
 
diff --git a/monai/networks/utils.py b/monai/networks/utils.py
index 90e6d2760d..a6b0699107 100644
--- a/monai/networks/utils.py
+++ b/monai/networks/utils.py
@@ -273,7 +273,7 @@ def pixelshuffle(
     dim, factor = spatial_dims, scale_factor
     input_size = list(x.size())
     batch_size, channels = input_size[:2]
-    scale_divisor = factor ** dim
+    scale_divisor = factor**dim
 
     if channels % scale_divisor != 0:
         raise ValueError(
diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index 45307ab918..f2875ebc72 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -184,9 +184,9 @@ def _add_noise(self, img: NdarrayOrTensor, mean: float, std: float):
         if isinstance(img, torch.Tensor):
             n1 = torch.tensor(self._noise1, device=img.device)
             n2 = torch.tensor(self._noise2, device=img.device)
-            return torch.sqrt((img + n1) ** 2 + n2 ** 2)
+            return torch.sqrt((img + n1) ** 2 + n2**2)
 
-        return np.sqrt((img + self._noise1) ** 2 + self._noise2 ** 2)
+        return np.sqrt((img + self._noise1) ** 2 + self._noise2**2)
 
     def __call__(self, img: NdarrayOrTensor, randomize: bool = True) -> NdarrayOrTensor:
         """
diff --git a/monai/transforms/smooth_field/array.py b/monai/transforms/smooth_field/array.py
index 61342b8941..f581687ea5 100644
--- a/monai/transforms/smooth_field/array.py
+++ b/monai/transforms/smooth_field/array.py
@@ -232,7 +232,7 @@ def __call__(self, img: NdarrayOrTensor, randomize: bool = True) -> NdarrayOrTen
         # everything below here is to be computed using the destination type (numpy, tensor, etc.)
 
         img = (img - img_min) / (img_rng + 1e-10)  # rescale to unit values
-        img = img ** rfield  # contrast is changed by raising image data to a power, in this case the field
+        img = img**rfield  # contrast is changed by raising image data to a power, in this case the field
 
         out = (img * img_rng) + img_min  # rescale back to the original image value range
 
diff --git a/monai/transforms/utils_create_transform_ims.py b/monai/transforms/utils_create_transform_ims.py
index 9fc9206fac..b096e1b93d 100644
--- a/monai/transforms/utils_create_transform_ims.py
+++ b/monai/transforms/utils_create_transform_ims.py
@@ -383,7 +383,7 @@ def get_images(data, is_label=False):
         # we might need to panel the images. this happens if a transform produces e.g. 4 output images.
         # In this case, we create a 2-by-2 grid from them. Output will be a list containing n_orthog_views,
         # each element being either the image (if num_samples is 1) or the panelled image.
-        nrows = int(np.floor(num_samples ** 0.5))
+        nrows = int(np.floor(num_samples**0.5))
         for view in range(num_orthog_views):
             result = np.asarray([d[view] for d in data])
             nindex, height, width = result.shape
diff --git a/tests/test_lmdbdataset.py b/tests/test_lmdbdataset.py
index b624e5c4e3..33f27ee4bc 100644
--- a/tests/test_lmdbdataset.py
+++ b/tests/test_lmdbdataset.py
@@ -57,7 +57,7 @@
         SimulateDelayd(keys=["image", "label", "extra"], delay_time=[1e-7, 1e-6, 1e-5]),
     ],
     (128, 128, 128),
-    {"pickle_protocol": 2, "lmdb_kwargs": {"map_size": 100 * 1024 ** 2}},
+    {"pickle_protocol": 2, "lmdb_kwargs": {"map_size": 100 * 1024**2}},
 ]
 
 TEST_CASE_6 = [
@@ -66,7 +66,7 @@
         SimulateDelayd(keys=["image", "label", "extra"], delay_time=[1e-7, 1e-6, 1e-5]),
     ],
     (128, 128, 128),
-    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 100 * 1024 ** 2}},
+    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 100 * 1024**2}},
 ]
 
 TEST_CASE_7 = [
@@ -75,7 +75,7 @@
         SimulateDelayd(keys=["image", "label", "extra"], delay_time=[1e-7, 1e-6, 1e-5]),
     ],
     (128, 128, 128),
-    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 2 * 1024 ** 2}},
+    {"db_name": "testdb", "lmdb_kwargs": {"map_size": 2 * 1024**2}},
 ]
 
 
diff --git a/tests/test_tile_on_grid.py b/tests/test_tile_on_grid.py
index 212f461c9a..09434de5e0 100644
--- a/tests/test_tile_on_grid.py
+++ b/tests/test_tile_on_grid.py
@@ -107,7 +107,7 @@ def make_image(
 
     tiles = np.stack(tiles_list, axis=0)
 
-    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count ** 2:
+    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count**2:
         tiles = tiles[np.argsort(tiles.sum(axis=(1, 2, 3)))]
 
     return imlarge, tiles
diff --git a/tests/test_tile_on_grid_dict.py b/tests/test_tile_on_grid_dict.py
index 609601f222..c6f35fe738 100644
--- a/tests/test_tile_on_grid_dict.py
+++ b/tests/test_tile_on_grid_dict.py
@@ -116,7 +116,7 @@ def make_image(
 
     tiles = np.stack(tiles_list, axis=0)
 
-    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count ** 2:
+    if (filter_mode == "min" or filter_mode == "max") and len(tiles) > tile_count**2:
         tiles = tiles[np.argsort(tiles.sum(axis=(1, 2, 3)))]
 
     return imlarge, tiles

From 19f0b4229a3cb237a37ff8e8de9fc6b6bae54984 Mon Sep 17 00:00:00 2001
From: yc7620 <yaniel.cabrera20@imperial.ac.uk>
Date: Tue, 22 Feb 2022 09:47:59 +0100
Subject: [PATCH 4/6] mapping via bitbucket operation

Signed-off-by: yc7620 <yaniel.cabrera20@imperial.ac.uk>
---
 monai/transforms/intensity/array.py | 69 ++++++++++++-----------------
 1 file changed, 28 insertions(+), 41 deletions(-)

diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index 7a50a528d0..86de06d223 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -2063,39 +2063,35 @@ class IntensityRemap(RandomizableTransform):
     pixel is replaced by a new values coming from an intensity remappping
     curve.
 
-    The remapping curve is created by uniformly sampling values from the
-    possible intensities for the input image and then adding a linear
+    The remapping curve is created by uniformly sampling values from the 
+    possible intensities for the input image and then adding a linear 
     component. The curve is the rescaled to the input image intensity range.
 
     Intended to be used as a means to data augmentation via:
     :py:class:`monai.transforms.RandIntensityRemap`.
 
-    Implementation is described in the work:
-    `Intensity augmentation for domain transfer of whole breast segmentation
-    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
-
-    Note: for large images the transform may be too slow to apply on the fly. 
+    Implementation is described in the work: 
+    `Intensity augmentation for domain transfer of whole breast segmentation 
+    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_. 
 
     Args:
-        kernel_size: window size for averaging operation for the remapping
+        kernel_size: window size for averaging operation for the remapping 
             curve.
         slope: slope of the linear component. Easiest to leave default value
             and tune the kernel_size parameter instead.
         return_map: set to True for the transform to return a dictionary version
             of the lookup table used in the intensity remapping. The keys
-            correspond to the old intensities, and the values are the new
+            correspond to the old intensities, and the values are the new 
             values.
     """
-
-    def __init__(self, kernel_size: int = 30, slope: float = 0.7, return_map: bool = False):
+    def __init__(self, kernel_size: int = 30, slope: float = 0.7):
 
         super().__init__()
 
         self.kernel_size = kernel_size
         self.slope = slope
-        self.return_map = return_map
 
-    def __call__(self, img: torch.Tensor) -> Union[torch.Tensor, Tuple[torch.Tensor, Mapping]]:
+    def __call__(self, img: torch.Tensor) -> torch.Tensor:
         """
         Args:
             img: image to remap.
@@ -2104,25 +2100,22 @@ def __call__(self, img: torch.Tensor) -> Union[torch.Tensor, Tuple[torch.Tensor,
         img = img.clone()
         # sample noise
         vals_to_sample = torch.unique(img).tolist()
-        noise = torch.from_numpy(self.R.choice(vals_to_sample, len(vals_to_sample) - 1 + self.kernel_size))
+        noise = torch.from_numpy(
+            self.R.choice(vals_to_sample, len(vals_to_sample) - 1 + self.kernel_size)
+        )
         # smooth
-        noise = torch.nn.AvgPool1d(self.kernel_size, stride=1)(noise.unsqueeze(0)).squeeze()
+        noise = AvgPool1d(self.kernel_size, stride = 1)(noise.unsqueeze(0)).squeeze()
         # add linear component
         grid = torch.arange(len(noise)) / len(noise)
-        noise += self.slope * grid
+        noise += self.slope * grid 
         # rescale
         noise = (noise - noise.min()) / (noise.max() - noise.min()) * img.max() + img.min()
 
         # intensity remapping function
-        map = dict(zip(vals_to_sample, noise.tolist()))
+        index_img = torch.bucketize(img, torch.tensor(vals_to_sample))
+        img = noise[index_img]
 
-        for key in map:
-            img[img == key] = map[key]
-
-        if self.return_map:
-            return img, map
-        else:
-            return img
+        return img
 
 
 class RandIntensityRemap(RandomizableTransform):
@@ -2131,32 +2124,29 @@ class RandIntensityRemap(RandomizableTransform):
     pixel is replaced by a new values coming from an intensity remappping
     curve.
 
-    The remapping curve is created by uniformly sampling values from the
-    possible intensities for the input image and then adding a linear
+    The remapping curve is created by uniformly sampling values from the 
+    possible intensities for the input image and then adding a linear 
     component. The curve is the rescaled to the input image intensity range.
 
-    Implementation is described in the work:
-    `Intensity augmentation for domain transfer of whole breast segmentation
-    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
+    Implementation is described in the work: 
+    `Intensity augmentation for domain transfer of whole breast segmentation 
+    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_. 
 
-    Note: for large images the transform may be too slow to apply on the fly.
-    
     Args:
         prob: probability of applying the transform.
-        kernel_size: window size for averaging operation for the remapping
+        kernel_size: window size for averaging operation for the remapping 
             curve.
         slope: slope of the linear component. Easiest to leave default value
             and tune the kernel_size parameter instead.
         channel_wise: set to True to treat each channel independently.
     """
-
     def __init__(self, prob: float = 0.1, kernel_size: int = 30, slope: float = 0.7, channel_wise: bool = True):
 
         RandomizableTransform.__init__(self, prob=prob)
         self.kernel_size = kernel_size
         self.slope = slope
         self.channel_wise = True
-
+        
     def __call__(self, img: torch.Tensor) -> torch.Tensor:
         """
         Args:
@@ -2166,12 +2156,9 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         if self._do_transform:
             if self.channel_wise:
                 img = torch.stack(
-                    [
-                        IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i])
-                        for i in range(len(img))
-                    ]
-                )
+                    [IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i]) 
+                                            for i in range(len(img))])
             else:
                 img = IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img)
-
-        return img
+  
+        return img
\ No newline at end of file

From b7809ae0dfc03616d071ac7a544559a1f687014c Mon Sep 17 00:00:00 2001
From: "pre-commit-ci[bot]"
 <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Date: Tue, 22 Feb 2022 08:54:44 +0000
Subject: [PATCH 5/6] [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci
---
 monai/transforms/intensity/array.py | 36 ++++++++++++++---------------
 1 file changed, 18 insertions(+), 18 deletions(-)

diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index d0ab8496d7..f4c697dfa8 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -2063,25 +2063,25 @@ class IntensityRemap(RandomizableTransform):
     pixel is replaced by a new values coming from an intensity remappping
     curve.
 
-    The remapping curve is created by uniformly sampling values from the 
-    possible intensities for the input image and then adding a linear 
+    The remapping curve is created by uniformly sampling values from the
+    possible intensities for the input image and then adding a linear
     component. The curve is the rescaled to the input image intensity range.
 
     Intended to be used as a means to data augmentation via:
     :py:class:`monai.transforms.RandIntensityRemap`.
 
-    Implementation is described in the work: 
-    `Intensity augmentation for domain transfer of whole breast segmentation 
-    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_. 
+    Implementation is described in the work:
+    `Intensity augmentation for domain transfer of whole breast segmentation
+    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
 
     Args:
-        kernel_size: window size for averaging operation for the remapping 
+        kernel_size: window size for averaging operation for the remapping
             curve.
         slope: slope of the linear component. Easiest to leave default value
             and tune the kernel_size parameter instead.
         return_map: set to True for the transform to return a dictionary version
             of the lookup table used in the intensity remapping. The keys
-            correspond to the old intensities, and the values are the new 
+            correspond to the old intensities, and the values are the new
             values.
     """
     def __init__(self, kernel_size: int = 30, slope: float = 0.7):
@@ -2107,7 +2107,7 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         noise = AvgPool1d(self.kernel_size, stride = 1)(noise.unsqueeze(0)).squeeze()
         # add linear component
         grid = torch.arange(len(noise)) / len(noise)
-        noise += self.slope * grid 
+        noise += self.slope * grid
         # rescale
         noise = (noise - noise.min()) / (noise.max() - noise.min()) * img.max() + img.min()
 
@@ -2124,17 +2124,17 @@ class RandIntensityRemap(RandomizableTransform):
     pixel is replaced by a new values coming from an intensity remappping
     curve.
 
-    The remapping curve is created by uniformly sampling values from the 
-    possible intensities for the input image and then adding a linear 
+    The remapping curve is created by uniformly sampling values from the
+    possible intensities for the input image and then adding a linear
     component. The curve is the rescaled to the input image intensity range.
 
-    Implementation is described in the work: 
-    `Intensity augmentation for domain transfer of whole breast segmentation 
-    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_. 
+    Implementation is described in the work:
+    `Intensity augmentation for domain transfer of whole breast segmentation
+    in MRI <https://ieeexplore.ieee.org/abstract/document/9166708>`_.
 
     Args:
         prob: probability of applying the transform.
-        kernel_size: window size for averaging operation for the remapping 
+        kernel_size: window size for averaging operation for the remapping
             curve.
         slope: slope of the linear component. Easiest to leave default value
             and tune the kernel_size parameter instead.
@@ -2146,7 +2146,7 @@ def __init__(self, prob: float = 0.1, kernel_size: int = 30, slope: float = 0.7,
         self.kernel_size = kernel_size
         self.slope = slope
         self.channel_wise = True
-        
+
     def __call__(self, img: torch.Tensor) -> torch.Tensor:
         """
         Args:
@@ -2156,9 +2156,9 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         if self._do_transform:
             if self.channel_wise:
                 img = torch.stack(
-                    [IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i]) 
+                    [IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i])
                                             for i in range(len(img))])
             else:
                 img = IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img)
-  
-        return img
\ No newline at end of file
+
+        return img

From 96819163dc470199a34d62d7a67b84c9a66e4c83 Mon Sep 17 00:00:00 2001
From: monai-bot <monai.miccai2019@gmail.com>
Date: Tue, 22 Feb 2022 11:11:07 +0000
Subject: [PATCH 6/6] [MONAI] python code formatting

Signed-off-by: monai-bot <monai.miccai2019@gmail.com>
---
 monai/transforms/intensity/array.py | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index f4c697dfa8..9419873fdd 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -2084,6 +2084,7 @@ class IntensityRemap(RandomizableTransform):
             correspond to the old intensities, and the values are the new
             values.
     """
+
     def __init__(self, kernel_size: int = 30, slope: float = 0.7):
 
         super().__init__()
@@ -2100,11 +2101,9 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         img = img.clone()
         # sample noise
         vals_to_sample = torch.unique(img).tolist()
-        noise = torch.from_numpy(
-            self.R.choice(vals_to_sample, len(vals_to_sample) - 1 + self.kernel_size)
-        )
+        noise = torch.from_numpy(self.R.choice(vals_to_sample, len(vals_to_sample) - 1 + self.kernel_size))
         # smooth
-        noise = AvgPool1d(self.kernel_size, stride = 1)(noise.unsqueeze(0)).squeeze()
+        noise = AvgPool1d(self.kernel_size, stride=1)(noise.unsqueeze(0)).squeeze()
         # add linear component
         grid = torch.arange(len(noise)) / len(noise)
         noise += self.slope * grid
@@ -2140,6 +2139,7 @@ class RandIntensityRemap(RandomizableTransform):
             and tune the kernel_size parameter instead.
         channel_wise: set to True to treat each channel independently.
     """
+
     def __init__(self, prob: float = 0.1, kernel_size: int = 30, slope: float = 0.7, channel_wise: bool = True):
 
         RandomizableTransform.__init__(self, prob=prob)
@@ -2156,8 +2156,11 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         if self._do_transform:
             if self.channel_wise:
                 img = torch.stack(
-                    [IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i])
-                                            for i in range(len(img))])
+                    [
+                        IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i])
+                        for i in range(len(img))
+                    ]
+                )
             else:
                 img = IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img)