diff --git a/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Efficient_RANSAC.h b/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Efficient_RANSAC.h
index 8b13c5333b5..a97ecb80df7 100644
--- a/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Efficient_RANSAC.h
+++ b/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Efficient_RANSAC.h
@@ -561,11 +561,13 @@ public:
           do {
 
             do {
-
-              first_sample = get_default_random()(
-                      static_cast<unsigned int>(m_num_available_points));
+              first_sample = get_default_random()(static_cast<unsigned int>(m_num_available_points));
             } while (m_shape_index[first_sample] != -1);
-            std::cerr << first_sample << std::endl;
+            std::cerr << "first sample: " << first_sample << std::endl;
+            std::cerr << "point index: " << *(m_input_iterator_first + first_sample) << std::endl;
+
+            auto point = get(m_point_pmap,
+                             *(m_input_iterator_first + first_sample));
 
             done =
                     drawSamplesFromCellContainingPoint(m_global_octree,
@@ -576,8 +578,6 @@ public:
                                                        m_shape_index,
                                                        required_samples);
 
-            std::cerr << m_shape_index[first_sample] << std::endl;
-
             if (callback && !callback(num_invalid / double(m_num_total_points)))
               return false;
 
@@ -1121,6 +1121,8 @@ private:
                                           const std::vector<int> &shapeIndex,
                                           std::size_t requiredSamples) {
 
+    std::cerr << "  point: " << p << std::endl;
+
     typedef typename Octree::Node Cell;
 
     const Cell *cur = node_containing_point(octree, p, level);
diff --git a/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Octree.h b/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Octree.h
index 7cf5cb259e7..dcaa922098c 100644
--- a/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Octree.h
+++ b/Shape_detection/include/CGAL/Shape_detection/Efficient_RANSAC/Octree.h
@@ -135,6 +135,7 @@ class Indexed_octree {
 
   typedef typename Traits::Input_range::iterator Input_iterator;
   typedef typename Traits::Point_map Point_map;
+  typedef typename Traits::FT FT;
   typedef std::vector<std::size_t> Input_range;
   typedef Point_map_to_indexed_point_map<Input_iterator, Point_map> Indexed_point_map;
 
@@ -177,7 +178,21 @@ public:
   void refine(double cluster_epsilon_for_max_level_recomputation = -1., std::size_t bucketSize = 2,
               std::size_t maxLevel = 10) {
 
-    // TODO: I need to find out what cluster_epsilon is used for
+    if (cluster_epsilon_for_max_level_recomputation > 0.) {
+
+      auto m_bBox = m_octree.bbox(m_octree.root());
+
+      FT bbox_diagonal = (FT) CGAL::sqrt(
+              (m_bBox.xmax() - m_bBox.xmin()) * (m_bBox.xmax() - m_bBox.xmin())
+              + (m_bBox.ymax() - m_bBox.ymin()) * (m_bBox.ymax() - m_bBox.ymin())
+              + (m_bBox.zmax() - m_bBox.zmin()) * (m_bBox.zmax() - m_bBox.zmin()));
+
+      maxLevel = std::size_t(std::log(bbox_diagonal
+                                      / cluster_epsilon_for_max_level_recomputation)
+                             / std::log(2.0));
+
+    }
+
     m_octree.refine(maxLevel, bucketSize);
   }