[Bf-blender-cvs] [0db119ad11e] gsoc-2018-many-light-sampling: Cycles: Updated SAOH with regularization factor

[Erik Englesson]

Commit: 0db119ad11e4738808fa9766829f8d0ddde18e5f
Author: Erik Englesson
Date:   Wed Jul 11 10:51:07 2018 +0200
Branches: gsoc-2018-many-light-sampling
https://developer.blender.org/rB0db119ad11e4738808fa9766829f8d0ddde18e5f

Cycles: Updated SAOH with regularization factor

This factor penalizes thin bounding boxes in
the tree construction.

===================================================================

M       intern/cycles/render/light_tree.cpp

===================================================================

diff --git a/intern/cycles/render/light_tree.cpp 
b/intern/cycles/render/light_tree.cpp
index 14a6f9e15c6..71eb71f4bf9 100644
--- a/intern/cycles/render/light_tree.cpp
+++ b/intern/cycles/render/light_tree.cpp
@@ -347,6 +347,7 @@ void LightTree::split_saoh(const BoundBox &centroidBbox,
        min_cost = std::numeric_limits<float>::max();
        min_bucket = -1;
 
+       const float extent_max = max3(centroidBbox.size());
        for (int dim = 0; dim < 3; ++dim){
 
                BucketInfo buckets[nBuckets];
@@ -358,12 +359,12 @@ void LightTree::split_saoh(const BoundBox &centroidBbox,
                        continue;
                }
 
-               const float invExtent = 1.0f / extent;
+               const float extent_inv = 1.0f / extent;
                for (unsigned int i = start; i < end; ++i)
                {
                        int bucket_id = (int)((float)nBuckets *
                                              (buildData[i].centroid[dim] - 
centroidBbox.min[dim]) *
-                                             invExtent);
+                                             extent_inv);
                        if (bucket_id == nBuckets) bucket_id = nBuckets - 1;
                        buckets[bucket_id].count++;
                        buckets[bucket_id].energy += buildData[i].energy;
@@ -391,6 +392,7 @@ void LightTree::split_saoh(const BoundBox &centroidBbox,
                        bcones_L.clear();
                        bcones_R.clear();
 
+                       /* L corresponds to all buckets up to and including i */
                        for (int j = 0; j <= i; ++j){
                                if (buckets[j].count != 0){
                                        energy_L += buckets[j].energy;
@@ -399,6 +401,7 @@ void LightTree::split_saoh(const BoundBox &centroidBbox,
                                }
                        }
 
+                       /* R corresponds to bucket i+1 and all after */
                        for (int j = i+1; j < nBuckets; ++j){
                                if (buckets[j].count != 0){
                                        energy_R += buckets[j].energy;
@@ -407,14 +410,24 @@ void LightTree::split_saoh(const BoundBox &centroidBbox,
                                }
                        }
 
-                       Orientation bcone_L = 
aggregate_bounding_cones(bcones_L);
-                       Orientation bcone_R = 
aggregate_bounding_cones(bcones_R);
-                       float M_Omega_L = calculate_cone_measure(bcone_L);
-                       float M_Omega_R = calculate_cone_measure(bcone_R);
-
-                       cost[i] = (energy_L*M_Omega_L*bbox_L.area() +
-                                  energy_R*M_Omega_R*bbox_R.area()) /
-                               (node_energy*node_M_Omega*node_bbox.area());
+                       /* eq. 2 */
+                       const Orientation bcone_L = 
aggregate_bounding_cones(bcones_L);
+                       const Orientation bcone_R = 
aggregate_bounding_cones(bcones_R);
+                       const float M_Omega_L     = 
calculate_cone_measure(bcone_L);
+                       const float M_Omega_R     = 
calculate_cone_measure(bcone_R);
+                       const float K             = extent_max * extent_inv;
+
+                       /* NOTE: in eq 2 they do not have the energy of the 
parent in the
+                        * denominator. Is this a typo?
+                        * Actually. The denominator does not affect the 
minimization since
+                        * it is just a constant scale factor for every cost 
being minimized.
+                        * The denominator might help with numerical issues 
though.
+                        *
+                        * TODO: When optimizing, could calc denominator once 
and take its
+                        * inverse once and then just multiply here. */
+                       cost[i] = K * (energy_L    * M_Omega_L    * 
bbox_L.area()  +
+                                      energy_R    * M_Omega_R    * 
bbox_R.area()) /
+                                     (node_energy * node_M_Omega * 
node_bbox.area());
                }
 
                /* update minimum cost, dim and bucket */

_______________________________________________
Bf-blender-cvs mailing list
Bf-blender-cvs@blender.org
https://lists.blender.org/mailman/listinfo/bf-blender-cvs