added auto global normalization option to minibatch

yapic_io/dataset.py

@@ -78,11 +78,27 @@ def image_dimensions(self, image_nr):
         '''
         return self.pixel_connector.image_dimensions(image_nr)
 
+
+    def _smallest_image_size_xy(self):
+        Z = 10
+        X = 500
+        Y = 500
+        for i in range(self.n_images):
+            size_z, size_x, size_y = self.image_dimensions(i)[-3:]
+            if size_z < Z:
+                Z = size_z
+            if size_x < X:
+                X = size_x
+            if size_y < Y:
+                Y = size_y
+        return (Z, X, Y)
+
+
     def pixel_statistics(self,
                          channels,
                          upper=99,
                          lower=1,
-                         tile_size_zxy=(1, 50, 50),
+                         tile_size_zxy=None,
                          n_tiles=1000):
         '''
         Performs random sampling of n tiles and calculates upper and lower
@@ -107,10 +123,11 @@ def pixel_statistics(self,
         [(lower_01, upper_01), (lower_02, upper_02), (lower_03, upper_03), ...]
         '''
 
-        tile_size_zxy = (1, 30, 30)
+        if tile_size_zxy is None:
+            tile_size_zxy = self._smallest_image_size_xy()
         percentiles = np.zeros((n_tiles, len(channels), 2))
-        msg = '\n\nCalculate global pixel statistics ({} tiles)...\n'.format(
-            n_tiles)
+        msg = ('\n\nCalculate global pixel statistics'
+               '({} tiles of size {})...\n').format(n_tiles, tile_size_zxy)
         sys.stdout.write(msg)
         for i in range(n_tiles):
             image_nr, pos_zxy = self._random_pos_izxy(None, tile_size_zxy)

yapic_io/minibatch.py

@@ -112,9 +112,10 @@ def set_normalize_mode(self, mode_str, minmax=None):
         self.normalize_mode = mode_str
 
         if mode_str == 'global':
-            assert minmax is not None, \
-                                 'normalization range (min, max) required'
 
+            if minmax is None:
+                # calculate upper and lower percentile automatically
+                minmax = self.dataset.pixel_statistics(self.channels)
             n_channels = self.dataset.pixel_connector.image_dimensions(0)[0]
 
             if len(minmax) == 2:

yapic_io/tests/test_training_batch.py

@@ -322,6 +322,22 @@ def test_normalize_multichannel(self):
         assert_array_equal(np.unique(pixels_normalized_zscore), [0])
         assert_array_equal(np.unique(pixels_normalized_local), [0])
 
+    def test_normalize_global_auto(self):
+
+        img_path = os.path.join(base_path, '../test_data/tiffconnector_1/im/')
+        label_path = os.path.join(base_path,
+                                  '../test_data/tiffconnector_1/labels/')
+        c = TiffConnector(img_path, label_path)
+        d = Dataset(c)
+
+        size = (1, 5, 4)
+        pad = (0, 0, 0)
+
+        m = TrainingBatch(d, size, padding_zxy=pad)
+        assert m.global_norm_minmax is None
+        m.set_normalize_mode('global')
+        assert len(m.global_norm_minmax) == 3
+
     def test_normalize_global_multichannel(self):
 
         img_path = os.path.join(base_path,