Merge ef5fdcb into 34dcb1b

deepcell/image_generators/__init__.py

@@ -174,6 +174,9 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
     elif transform == 'centroid':
         erosion = kwargs.pop('erosion_width', 0)
         disk_size = kwargs.pop('disk_size', 4)
+        alpha = kwargs.pop('alpha', 0.1)
+        auto_alpha = kwargs.pop('auto_alpha', False)
+
         if data_format == 'channels_first':
             y_transform = np.zeros(tuple([y.shape[0]] + list(y.shape[2:])))
         else:
@@ -190,7 +193,8 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
             else:
                 mask = y[batch, ..., 0]
 
-            y_transform[batch] = _transform(mask, erosion)
+            y_transform[batch] = _transform(mask, erosion_width=erosion,
+                                            alpha=alpha, auto_alpha=auto_alpha)
 
         y_transform = np.expand_dims(y_transform, axis=-1)
 

deepcell/utils/transform_utils.py

@@ -234,12 +234,20 @@ def centroid_transform_continuous_2d(mask, erosion_width=None, alpha=0.1):
     Args:
         mask (numpy.array): a label mask (y data)
         erosion_width (int): number of pixels to erode edges of each labels
-        alpha (float): coefficent to reduce the magnitude of the distance value.
+        alpha (float, str): coefficent to reduce the magnitude of the distance
+            value. If 'auto', determines alpha for each cell based on the cell
+            area. Defaults to 0.1.
 
     Returns:
         numpy.array: a mask of same shape as input mask,
             with each label being a distance class from 1 to bins
     """
+
+    # Check input to alpha
+    if isinstance(alpha, str):
+        if alpha.lower() != 'auto':
+            raise ValueError('alpha must be set to "auto"')
+
     mask = np.squeeze(mask)
     mask = erode_edges(mask, erosion_width)
 
@@ -253,7 +261,14 @@ def centroid_transform_continuous_2d(mask, erosion_width=None, alpha=0.1):
         coords = prop.coords
         center = prop.weighted_centroid
         distance_to_center = np.sum((coords - center) ** 2, axis=1)
-        center_transform = 1 / (1 + alpha * distance_to_center)
+
+        # Determine alpha to use
+        if str(alpha).lower() == 'auto':
+            _alpha = 1 / np.sqrt(prop.area)
+        else:
+            _alpha = float(alpha)
+
+        center_transform = 1 / (1 + _alpha * distance_to_center)
         coords_x = coords[:, 0]
         coords_y = coords[:, 1]
         inner_distance[coords_x, coords_y] = center_transform
@@ -267,7 +282,8 @@ def centroid_transform_continuous_movie(mask, erosion_width=None, alpha=0.1):
     Args:
         mask (numpy.array): a label mask (y data)
         erosion_width (int): number of pixels to erode edges of each labels
-        alpha (float): coefficent to reduce the magnitude of the distance value.
+        alpha (float): coefficent to reduce the magnitude of the distance
+            value.
 
     Returns:
         numpy.array: a mask of same shape as input mask,