--- a
+++ b/rocaseg/preproc/custom.py
@@ -0,0 +1,55 @@
+import logging
+import numpy as np
+
+
+logging.basicConfig()
+logger = logging.getLogger('preprocessing_custom')
+logger.setLevel(logging.DEBUG)
+
+
+class Normalize:
+    def __init__(self, mean, std):
+        self.mean = mean
+        self.std = std
+
+    def __call__(self, img, mask=None):
+        img = img.astype(np.float32)
+        img = (img - self.mean) / self.std
+
+        if mask is not None:
+            mask = mask.astype(np.float32)
+        return img, mask
+
+
+class UnNormalize:
+    def __init__(self, mean, std):
+        self.mean = mean
+        self.std = std
+
+    def __call__(self, *args):
+        return [(a * self.std + self.mean) for a in args]
+
+
+class PercentileClippingAndToFloat:
+    """Change the histogram of image by doing global contrast normalization."""
+    def __init__(self, cut_min=0.5, cut_max=99.5):
+        """
+        cut_min - lowest percentile which is used to cut the image histogram
+        cut_max - highest percentile
+        """
+        self.cut_min = cut_min
+        self.cut_max = cut_max
+
+    def __call__(self, img, mask=None):
+        img = img.astype(np.float32)
+        lim_low, lim_high = np.percentile(img, [self.cut_min, self.cut_max])
+        img = np.clip(img, lim_low, lim_high)
+
+        img -= lim_low
+        img /= img.max()
+
+        img = img.astype(np.float32)
+        if mask is not None:
+            mask = mask.astype(np.float32)
+
+        return img, mask