Add clip into undo_norm_range01 as it's the expected value range

biapy/data/pre_processing.py

@@ -846,7 +846,7 @@ def norm_range01(x, dtype=np.float32):
     norm_steps = {}
     norm_steps['orig_dtype'] = x.dtype
 
-    if x.dtype == np.uint8 or x.dtype == torch.uint8:
+    if x.dtype in [np.uint8, torch.uint8]:
         x = x/255
         norm_steps['div'] = 1
     else:
@@ -867,20 +867,27 @@ def norm_range01(x, dtype=np.float32):
 
 def undo_norm_range01(x, xnorm):
     if 'div' == xnorm['type']:
-        x = (x*255)
-        if isinstance(x, np.ndarray): 
-            x = x.astype(np.uint8)
+        # Prevent values go outside expected range 
+        if isinstance(x, np.ndarray):
+            x = np.clip(x, 0, 1) 
         else:
-            x = x.to(torch.uint8)
-    reductions = [key for key, value in xnorm.items() if 'reduced' in key.lower()]
-    if len(reductions)>0:
-        reductions = reductions[0]
-        reductions = reductions.replace('reduced_','')
-        x = (x*65535)
-        if isinstance(x, np.ndarray): 
-            x = x.astype(eval("np.{}".format(reductions) ))
+            x = torch.clamp(x, 0, 1)
+        if 'div' in xnorm:
+            x = (x*255)
+            if isinstance(x, np.ndarray): 
+                x = x.astype(np.uint8)
+            else:
+                x = x.to(torch.uint8)
         else:
-            x = x.to(eval("torch.{}".format(reductions) ))
+            reductions = [key for key, value in xnorm.items() if 'reduced' in key.lower()]
+            if len(reductions)>0:
+                reductions = reductions[0]
+                reductions = reductions.replace('reduced_','')
+                x = (x*65535)
+                if isinstance(x, np.ndarray): 
+                    x = x.astype(eval("np.{}".format(reductions) ))
+                else:
+                    x = x.to(eval("torch.{}".format(reductions) ))            
     return x
 
 def reduce_dtype(x, x_min, x_max, out_min=0, out_max=1, out_type=np.float32):