Dynamic terrain exaggeration #9603

IanLilleyT · 2021-06-09T01:16:56Z

dynamic_terrain_exaggeration.mov

This PR adds dynamic exaggeration to terrain, an enhancement over the fixed terrain exaggeration that CesiumJS has had since 2015. This property is controlled by Globe.terrainExaggeration which can be changed as often as you want. There's also a new property called Globe.terrainExaggerationRelativeHeight which is the height relative to which terrain is exaggerated, meaning terrain above that height will stretch upwards, and terrain below that height will stretch downwards. Globe.terrainExaggeration is 1.0 by default and Globe.terrainExaggerationRelativeHeight is 0.0 by default.

The old ways of setting terrainExaggeration have been deprecated and will be removed in 1.85: #9626

It works by storing a f32vec3 geodetic surface normal per-vertex and adding it to the relative-to-center position in the shader, which is about as good as we can get accuracy-wise without 64-bit emulation or some other slow path.

float newHeight = (height - relativeHeight) * exaggeration + relativeHeight;
positionRTC += geodeticSurfaceNormal * (newHeight - height);
... convert to world coordinates from positionRTC

The geodetic surface normals are only stored when exaggeration is not 1.0. If you go from 1.0 to another number, they will be computed only the fly. If you go from not 1.0 to 1.0, they will be removed. One downside to this PR is more memory is needed for dynamic terrain exaggeration compared to fixed terrain exaggeration. This is because fixed terrain exaggeration baked the exaggeration into the geometry, so it didn't need to store geodetic surface normals. To get a sense of how much extra memory is used, see the screenshots below, especially the TerrainMesh category. Here is comes out to ~30% extra memory.

1.0 exaggeration

1.01 exaggeration

And back to 1.0 exaggeration. This is the same amount of TerrainMesh memory as the first image.

Also, this PR fixes #7580 - problem with exaggerated normals getting washed out. It looks better now:

Normals are exaggerated in the shader by scaling the "rejection" between the original vertex normal and the geodetic surface normal. So if the slope is originally 45 degrees and gets exaggerated by 100X, the exaggerated normal is going to be mostly perpendicular to the geodetic surface normal. It seems to work a lot better than the old way and is fairly cheap to calculate and is useful for lighting at the very least.

vec3 projection = dot(normalMC, geodeticSurfaceNormal) * geodeticSurfaceNormal
vec3 rejection = normalMC - projection;
normalMC = normalize(projection + rejection * exaggeration);

Finally, I whipped up a test to see how much floating point error is introduced by moving the exaggeration code to the shader instead of baking into the geometry. Baking has 64-bit precision and shader has 32-bit precision. The most error I recorded was under 10cm. The sample code below came out with an error of 0.063323 meters, with the inputs:

Height: -0.294904
RTC: (-757482.432647, -723523.092327, -279114.790856)
Exaggeration: 3.248070
Relative Height: 863.790587

RTC is the most important factor when determining error. When the RTC is forced to be between -10,000 and +10,000, the maximum error is 0.001172m

See full results

I've tested with:

2D/CV
HeightReference.CLAMP_TO_GROUND
Globe.pick
ScreenSpaceCameraController

It's always possible I've missed a feature to test. Let me know if so.

This PR doesn't fix problems related to ground primitive shadow volumes not clamping to the ground properly: #8480. So if you exaggerate terrain there could still be problems there.

cesium-concierge · 2021-06-09T01:16:59Z

Thanks for the pull request @IanLilleyT!

✔️ Signed CLA found.
❔ Unit tests were not updated.
- Make sure you've updated tests to reflect your changes, added tests for any new code, and ran the code coverage tool.

Reviewers, don't forget to make sure that:

Cesium Viewer works.
Works in 2D/CV.
Works (or fails gracefully) in IE11.

IanLilleyT · 2021-06-09T12:50:56Z

Source/Core/HeightmapTerrainData.js

@@ -805,23 +812,19 @@ function interpolateMeshHeight(
  northInteger = height - 1 - northInteger;

  var southwestHeight =
-    (encoding.decodeHeight(buffer, southInteger * width + westInteger) /
-      exaggeration -
+    (encoding.decodeHeight(buffer, southInteger * width + westInteger) -


Exaggeration used to be baked into the TerrainMesh geometry and had to be undone here. Now there's no need for that.

IanLilleyT · 2021-06-09T12:52:22Z

Source/Core/HeightmapTessellator.js


  var isGeographic = defaultValue(options.isGeographic, true);
  var ellipsoid = defaultValue(options.ellipsoid, Ellipsoid.WGS84);

  var oneOverGlobeSemimajorAxis = 1.0 / ellipsoid.maximumRadius;

-  var nativeRectangle = options.nativeRectangle;
+  var nativeRectangle = Rectangle.clone(options.nativeRectangle);


The worker strips the functions from these inputs, so clone to turn them into real Rectangle again.

IanLilleyT · 2021-06-09T12:55:15Z

Source/Core/QuantizedMeshTerrainData.js

-      OrientedBoundingBox.clone(result.orientedBoundingBox),
-      that._orientedBoundingBox
-    );
+    var boundingSphere = that._boundingSphere;


The create mesh worker doesn't return bounding sphere and obb any more. Now, GlobeSurfaceTileProvider regenerates bounding volumes when exaggeration changes.

IanLilleyT · 2021-06-09T12:56:25Z

Source/Core/TerrainEncoding.js

@@ -23,25 +24,32 @@ var SHIFT_LEFT_12 = Math.pow(2.0, 12.0);
 * @alias TerrainEncoding
 * @constructor
 *
+ * @param {Cartesian3} center The center point of the vertices.


Before it used the AxisAlignedBoundingBox.center as the RTC center, which is not necessarily the best possible value, and was also unclear.

IanLilleyT · 2021-06-09T12:57:43Z

Source/Core/TerrainEncoding.js

+  this._calculateStrideAndOffsets();
+  var newStride = this.stride;
+
+  for (var index = 0; index < vertexCount; index++) {


This code assumes geodeticSurfaceNormal is the last vertex attribute. This could probably use a refactor when we add another vertex attribute after

IanLilleyT · 2021-06-09T13:01:21Z

Source/Core/TerrainMesh.js

@@ -45,7 +44,6 @@ function TerrainMesh(
  vertexStride,
  orientedBoundingBox,
  encoding,


TerrainMesh doesn't need exaggeration any more. We send the un-exaggerated positions to the GPU and scale dynamically. In JS we would call TerrainEncoding.getExaggeratedPosition

IanLilleyT · 2021-06-09T13:03:23Z

Source/Scene/FrameState.js

+   * @type {Number}
+   * @default 0.0
+   */
+  this.terrainExaggerationRelativeHeight = 0.0;


FrameState still needs exaggeration values because GlobeSurfaceTileProvider doesn't have access to the Globe, but does have access to FrameState. This is used when checking if the exaggeration has changed between frames so that geodetic surface normals can be added or removed.

IanLilleyT · 2021-06-09T13:04:24Z

Source/Scene/Globe.js

@@ -672,7 +692,7 @@ Globe.prototype.pickWorldCoordinates = function (
      );
    } else if (defined(surfaceTile.renderedMesh)) {
      BoundingSphere.clone(
-        surfaceTile.renderedMesh.boundingSphere3D,
+        surfaceTile.tileBoundingRegion.boundingSphere,


The mesh's boundingSphere3D is not exaggerated, so it can't be used. The TileBoundingRegion.boundingSphere has the updated bounding volumes.

IanLilleyT · 2021-06-09T13:06:59Z

Source/Scene/GlobeSurfaceTile.js

@@ -437,6 +511,7 @@ function processTerrainStateMachine(
        frameState,
        terrainProvider,
        imageryLayerCollection,
+        vertexArraysToDestroy,


A missing param that was never caught....

IanLilleyT · 2021-06-09T13:11:01Z

Source/Scene/QuadtreePrimitive.js

@@ -1514,38 +1518,6 @@ function updateHeights(primitive, frameState) {
          data.callback(position);
          data.level = tile.level;
        }
-      } else if (tile.level === data.level) {


This used to remove the height clamp callback from the object when the terrain tile beneath it was refined as far as possible, possibly as an optimization since the height was never going to get more accurate or change. This code has been deleted because terrain exaggeration can happen at any time, so the callback needs to stick around.

CHANGES.md

Source/Core/GoogleEarthEnterpriseTerrainData.js

lilleyse · 2021-06-17T21:52:19Z

Source/Scene/GlobeSurfaceTileProvider.js

+        }
+      } else {
+        // this tile may come into view at a later time so keep the loop active
+        continueProcessing = true;


What if the tile is never seen again? Does the loop happen every frame forever until the tile is unloaded?

In the sandcastle if I click "Remove Exaggeration" I see "here" get printed continuously.

quadtree.forEachLoadedTile(function (tile) { console.log("here"); ... }

Should each tile have a dirty frame so that it knows to add or remove its geodetic surface normals?

After offline discussion forEachRenderedTile seems to work.

lilleyse · 2021-06-17T23:29:46Z

Source/Scene/Scene.js

@@ -1432,15 +1438,32 @@ Object.defineProperties(Scene.prototype, {
  },

  /**
-   * Gets the scalar used to exaggerate the terrain.
+   * Gets or sets the scalar used to exaggerate the terrain.
   * @memberof Scene.prototype
   * @type {Number}
   * @readonly


No longer @readonly (though it is going away so it doesn't really matter)

Source/Widgets/CesiumWidget/CesiumWidget.js

lilleyse · 2021-06-18T00:06:09Z

Fog culling is really severe with strong exaggeration. Out of scope for this PR. CC #8454

Sandcastle

… unrendered straggler tiles

… going to be removed soon

IanLilleyT · 2021-06-18T15:37:21Z

@lilleyse This is ready for another review

lilleyse · 2021-06-18T19:40:01Z

The updates look good. eslint and tests pass locally.

Dynamic terrain exaggeration

ad9da87

Fixed PR number in CHANGES.md

25405cb

IanLilleyT commented Jun 9, 2021

View reviewed changes

Merge branch 'master' into dynamicTerrainExaggeration

f55bac2

lilleyse reviewed Jun 17, 2021

View reviewed changes

IanLilleyT mentioned this pull request Jun 18, 2021

Remove Scene.terrainExaggeration, Scene.options.terrainExaggeration, CesiumWidget.options.terrainExaggeration, and Viewer.options.terrainExaggeration in 1.85 #9626

Closed

IanLilleyT added 3 commits June 18, 2021 11:35

added terrainExaggeration deprecation warning for CesiumWidget

0949896

using forEachRenderedTile instead of forEachLoadedTile to not process…

d2ba60a

… unrendered straggler tiles

Removed readonly flag for Scene.terrainExaggeration, even though it's…

253fada

… going to be removed soon

Merge branch 'master' into dynamicTerrainExaggeration

131280f

lilleyse merged commit 490e522 into master Jun 18, 2021

lilleyse deleted the dynamicTerrainExaggeration branch June 18, 2021 19:40

IanLilleyT mentioned this pull request Aug 29, 2021

Remove deprecated ways of setting terrain exaggeration and a bugfix #9766

Merged

jjhembd mentioned this pull request Nov 29, 2023

Vertical exaggeration for 3D Tiles #11655

Merged

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Dynamic terrain exaggeration #9603

Dynamic terrain exaggeration #9603

IanLilleyT commented Jun 9, 2021 •

edited

Loading

cesium-concierge commented Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021

IanLilleyT Jun 9, 2021 •

edited

Loading

lilleyse Jun 17, 2021

lilleyse Jun 17, 2021

lilleyse Jun 17, 2021

lilleyse commented Jun 18, 2021

IanLilleyT commented Jun 18, 2021

lilleyse commented Jun 18, 2021

Dynamic terrain exaggeration #9603

Dynamic terrain exaggeration #9603

Conversation

IanLilleyT commented Jun 9, 2021 • edited Loading

cesium-concierge commented Jun 9, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

IanLilleyT Jun 9, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lilleyse commented Jun 18, 2021

IanLilleyT commented Jun 18, 2021

lilleyse commented Jun 18, 2021

IanLilleyT commented Jun 9, 2021 •

edited

Loading

IanLilleyT Jun 9, 2021 •

edited

Loading