Use the area weighted inverse normal for GUI Cameras.

We will now calculate the area-weighted normal of all polygons in the
GUI page, then flip it to create the direction that the GUI camera
should face. This will, theoretically, be more intuitive to artists in
that the GUI camera should always point at what they've modelled instead
of imposing arbitrary rules on the coordinate system of GUI objects.

korman/exporter/gui.py

@@ -20,6 +20,7 @@ import mathutils
 
 from contextlib import contextmanager, ExitStack
 import itertools
+import math
 from PyHSPlasma import *
 from typing import *
 import weakref
@@ -84,27 +85,28 @@ class GuiConverter:
         if not objects:
             raise ExportError("No objects specified for GUI Camera generation.")
 
-        class ObjArea(NamedTuple):
-            obj: bpy.types.Object
-            area: float
-
-
+        # Generally, GUIs are flat planes. However, we are not Cyan, so artists cannot walk down
+        # the hallway to get smacked on the knuckles by programmers. This means that they might
+        # give us some three dimensional crap as a GUI. Therefore, to come up with a camera matrix,
+        # we'll use the average area-weighted inverse normal of all the polygons they give us. That
+        # way, the camera *always* should face the GUI as would be expected.
         remove_mesh = bpy.data.meshes.remove
-        obj_areas: List[ObjArea] = []
+        avg_normal = mathutils.Vector()
         for i in objects:
             mesh = i.to_mesh(bpy.context.scene, True, "RENDER", calc_tessface=False)
             with helpers.TemporaryObject(mesh, remove_mesh):
                 utils.transform_mesh(mesh, i.matrix_world)
-                obj_areas.append(
-                    ObjArea(i, sum((polygon.area for polygon in mesh.polygons)))
-                )
-        largest_obj = max(obj_areas, key=lambda x: x.area)
-
-        # GUIs are generally flat planes, which, by default in Blender, have their normal pointing
-        # in the localspace z axis. Therefore, what we would like to do is have a camera that points
-        # at the localspace -Z axis. Where up is the localspace +Y axis. We are already pointing at
-        # -Z with +Y being up from what I can tel. So, just use this matrix.
-        mat = largest_obj.obj.matrix_world.to_3x3()
+                for polygon in mesh.polygons:
+                    avg_normal += (polygon.normal * polygon.area)
+        avg_normal.normalize()
+        avg_normal *= -1.0
+
+        # From the inverse area weighted normal we found above, get the rotation from the up axis
+        # (that is to say, the +Z axis) and create our rotation matrix.
+        axis = mathutils.Vector((avg_normal.x, avg_normal.y, 0.0))
+        axis.normalize()
+        angle = math.acos(avg_normal.z)
+        mat = mathutils.Matrix.Rotation(angle, 3, axis)
 
         # Now, we know the rotation of the camera. Great! What we need to do now is ensure that all
         # of the objects in question fit within the view of a 4:3 camera rotated as above. Blender
@@ -120,10 +122,10 @@ class GuiConverter:
             camera.data.lens_unit = "FOV"
 
             # Get all of the bounding points and make sure they all fit inside the camera's view frame.
-            bound_boxes = (
+            bound_boxes = [
                 obj.matrix_world * mathutils.Vector(bbox)
                 for obj in objects for bbox in obj.bound_box
-            )
+            ]
             co, _ = camera.camera_fit_coords(
                 scene,
                 # bound_box is a list of vectors of each corner of all the objects' bounding boxes;
@@ -132,11 +134,24 @@ class GuiConverter:
                 list(itertools.chain.from_iterable(bound_boxes))
             )
 
-            # Calculate the distance from the largest object to the camera. The scale we were given
-            # will be used to push the camera back in the Z+ direction of that object by scale times.
-            bvh = mathutils.bvhtree.BVHTree.FromObject(largest_obj.obj, scene)
-            loc, normal, index, distance = bvh.find_nearest(co)
-            co += normal * distance * (scale - 1.0)
+            # This generates a list of 6 faces per bounding box, which we then flatten out and pass
+            # into the BVHTree constructor. This is to calculate the distance from the camera to the
+            # "entire GUI" - which we can then use to apply the scale given to us.
+            if scale != 1.0:
+                bvh = mathutils.bvhtree.BVHTree.FromPolygons(
+                    bound_boxes,
+                    list(itertools.chain.from_iterable(
+                        [(i + 0, i + 1, i + 5, i + 4),
+                         (i + 1, i + 2, i + 5, i + 6),
+                         (i + 3, i + 2, i + 6, i + 7),
+                         (i + 0, i + 1, i + 2, i + 3),
+                         (i + 0, i + 3, i + 7, i + 4),
+                         (i + 4, i + 5, i + 6, i + 7),
+                        ] for i in range(0, len(bound_boxes), 8)
+                    ))
+                )
+                loc, normal, index, distance = bvh.find_nearest(co)
+                co += normal * distance * (scale - 1.0)
 
             # ...
             mat.resize_4x4()