final upload

+import cv2
+import random
+import numpy
+from PIL import Image 
+
+class randomScale(object):
+  def __init__(self, dist):
+    self.dist = dist
+    
+
+  def __call__(self, img):
+    #if random.randrange(0,4) == 0:
+    #  return img
+    self.d = random.randrange(0,2)
+    img = numpy.array(img)
+    img_dtype = img.dtype
+    h, w, _ = img.shape
+    x = w if self.d == 0 else h 
+
+    n = int(x / self.dist) - 1 # number of lines
+    lines = [i * self.dist for i in range(n + 2)] # initial idx of each line
+    
+    for i in range(1, len(lines) - 1):
+        s, m, e = lines[i - 1], lines[i], lines[i + 1] # start, middle, end Lines
+        plusminus = random.randrange(0,2) # 0 = plus, 1 = minus
+        scale_range = random.randrange(int(self.dist / 4),int(self.dist / 2) + 1) # number of pixels for scaling
+        scale_range *= 1 if plusminus == 0 else -1
+
+        # scaling 
+        firstB = cv2.resize(img[0:h, s:m], (m - s + scale_range, h)) if self.d == 0 else cv2.resize(img[s:m, 0:w], (w, m - s + scale_range))
+        secondB = cv2.resize(img[0:h, m:e], (e - m - scale_range, h)) if self.d == 0 else cv2.resize(img[m:e, 0:w], (w, e - m - scale_range))
+
+        # applying scaled part to original img
+        if self.d == 0:
+            temp = cv2.hconcat([firstB,secondB])
+            img[0:h, s:e] = temp
+        else:
+            temp = cv2.vconcat([firstB,secondB])
+            img[s:e, 0:w] = temp
+    
+    img = img.astype(img_dtype)
+    return Image.fromarray(img)
+
+  def __repr__(self):
+    return self.__class__.__name__+'()'
+    
+