Allow using kmeans++ for initialization

src/mesh_segmentation/__init__.py

@@ -63,6 +63,11 @@ class MeshSegmentation(bpy.types.Operator):
                       "without sacrificing quality. 'Dense' can be tried as "
                       "fallback. Default",
         default = 'sparse')
+    kmeans_init: bpy.props.EnumProperty(name = "k-means initialization",
+        items = [('liu_zhang', "Liu & Zhang", "Initialization by Liu & Zhang"),
+                 ('kmeans++', "k-means++", "Initialization from k-means++")],
+        description = "Method to use for initializing centroids for k-means.",
+        default = 'liu_zhang')
 
     def execute(self, context):
         """Executes the segmentation"""
@@ -75,7 +80,8 @@ def execute(self, context):
                                       k = self.k,
                                       coefficients = (self.delta, self.eta),
                                       action = getattr(actions, self.action),
-                                      ev_method = self.ev_method)
+                                      ev_method = self.ev_method,
+                                      kmeans_init = self.kmeans_init)
             return {'FINISHED'}
 
     def invoke(self, context, event):

src/mesh_segmentation/segmentation.py

@@ -148,7 +148,7 @@ def _initial_guess(Q, k):
     return chosen
 
 
-def segment_mesh(mesh, k, coefficients, action, ev_method):
+def segment_mesh(mesh, k, coefficients, action, ev_method, kmeans_init):
     """Segments the given mesh into k clusters and performs the given
     action for each cluster
     """
@@ -175,15 +175,18 @@ def segment_mesh(mesh, k, coefficients, action, ev_method):
     # normalize each row to unit length
     V /= numpy.linalg.norm(V, axis=1)[:,None]
 
-    print("mesh_segmentation: Preparing kmeans...")
-    # compute association matrix
-    Q = V.dot(V.transpose())
-    # compute initial guess for clustering
-    initial_centroids = _initial_guess(Q, k)
-
-    print("mesh_segmentation: Applying kmeans...")
-    # apply kmeans
-    _, idx = scipy.cluster.vq.kmeans2(V, V[initial_centroids,:], iter=50)
+    if kmeans_init == 'kmeans++':
+        print("mesh_segmentation: Applying kmeans...")
+        _, idx = scipy.cluster.vq.kmeans2(V, k, minit='++', iter=50)
+    else:
+        print("mesh_segmentation: Preparing kmeans...")
+        # compute association matrix
+        Q = V.dot(V.transpose())
+        # compute initial guess for clustering
+        initial_centroids = _initial_guess(Q, k)
+
+        print("mesh_segmentation: Applying kmeans...")
+        _, idx = scipy.cluster.vq.kmeans2(V, V[initial_centroids,:], iter=50)
 
     print("mesh_segmentation: Done clustering!")
     # perform action with the clustering result