Oriented bounding box & frustum culling optimizations #2782

kainino0x · 2015-06-03T16:26:21Z

(Do not merge. Will be tested with 3D buildings first.)

In review & editing.

Further testing of rectangles which span the equator
Overall (rendering and network) performance benchmarking. Update the TODO in CHANGES.md.

Future:

Hook up to heightmapped terrain (and flat terrain, if that's separate); not just quantized meshes
Hook up to 2D and Columbus view rendering modes

pjcozzi · 2015-06-03T17:54:25Z

Source/Core/Matrix3.js

+     * @type {Matrix3}
+     * @constant
+     */
+    Matrix3.ZERO = freezeObject(new Matrix3(0.0, 0.0, 0.0,


For consistency, add this to Matrix2 and Matrix4. Mention these in CHANGES.md.

Where in CHANGES.md should I put it? Since I'm assuming the 1.10 section is final.

Make a new section for 1.11.

pjcozzi · 2015-06-03T18:15:32Z

This fails JSHint, see https://travis-ci.org/AnalyticalGraphicsInc/cesium/builds/65273151

pjcozzi · 2015-06-03T18:16:38Z

Folks - this is the start of a pull request for more exact view frustum culling so we do not plan to merge this until we have it hooked up to terrain.

kainino0x · 2015-06-03T19:17:16Z

Source/Core/OrientedBoundingBox.js

+    };
+
+    /**
+     * XXX Temporary compatibility function - remove after rebasing ono the rename from intersect to intersectPlane.


TODO for self: remove when rebasing onto global rename from intersect to intersectPlane

pjcozzi · 2015-06-03T19:32:51Z

Should I do the intersect -> intersectPlane rename in a separate PR and rebase onto that?

It's fine to include it in this PR.

kainino0x · 2015-06-05T17:38:39Z

TODO: Since we're changing intersect to intersectPlane anyway, consider replacing the Cartesian4 plane argument with an actual Plane plane.

pjcozzi · 2015-06-05T17:56:12Z

+1 on changing to Plane. This code was been written before Plane.

pjcozzi · 2015-06-05T17:56:33Z

Just double check how wide of a breaking change it becomes.

kainino0x · 2015-06-09T15:41:32Z

@pjcozzi CullingVolume exposes a {Cartesian4[]} field that it uses internally with .intersectPlane. Nothing public needs to change except for .intersectPlane now having a different signature than .intersect. It seems that it would behoove us to add Plane.fromCartesian4 and probably Plane.UNIT_X/Y/Z or similar.

pjcozzi · 2015-06-09T16:59:53Z

All sounds good.

This affects bounding volumes AxisAlignedBoundingBox and BoundingSphere.

kainino0x · 2015-06-17T21:51:27Z

This simple fix seems to work. I will have to do a bit more testing to see if I can still find similar issues. I did at one point have very similar changes in this file, but I probably didn't implement them correctly (I removed them because they weren't fixing this problem). This change was very small, though, since I had already wired it up to work last time I tried.

kainino0x · 2015-06-18T21:55:55Z

There have been a few changes that should be reviewed but other than these I don't know of anything I might change.

I added an adjustment to expand the bounding box to cover the tile skirt. This seems like something we should have, although it seems to sometimes pretty significantly increase the number of tiles rendered (279 to 298 compared to 330 for the sphere in one case). Any comments?

kainino0x · 2015-06-18T22:06:17Z

I'm working on a blog post draft here: https://github.com/kainino0x/draft-obb-blog-post

kring · 2015-06-18T23:19:49Z

I added an adjustment to expand the bounding box to cover the tile skirt.

It shouldn't be necessary to account for the skirts, since they're only intended to fill the vertical space between tiles. It's hard to imagine a situation where only the skirt is visible but the rest of the tile is not. You haven't seen such a case have you?

pjcozzi · 2015-06-19T16:32:43Z

@kainino0x and I talked offline, we are not going to include the skirts, but we'll comment it in the code incase we ever see cracks.

pjcozzi · 2015-06-19T16:40:58Z

I'm working on a blog post draft here: https://github.com/kainino0x/draft-obb-blog-post

Looks great. I will do some minor edits and publish next week. Is there anything else you wanted to add/change?

kainino0x · 2015-06-19T16:46:37Z

I'll change the numbers/screenshots a bit to reflect the removal of the skirt change. I'll push that to the repo, so you'll have it when you go to do the editing.

kainino0x · 2015-06-19T17:36:47Z

279 to 298 compared to 330 for the sphere in one case

For the record, the case in which the skirts hurt the performance was this one: "Toward horizon, low altitude".

kainino0x · 2015-06-19T17:48:58Z

In the blog post, should I mention why the bounding box appears not to contain the tile, because of the skirt sticking out?

…y semantics

pjcozzi · 2015-06-19T19:40:45Z

Yes, I think that is an interesting point to make: the box doesn't full bound the geometry, but, in practice, we don't expect it to generate artifacts since the view position/direction would never be such that it would matter.

pjcozzi · 2015-06-21T16:27:29Z

@kainino0x let me know when the blog post is ready and I will copyedit and publish this week.

Also, let me know this is ready to merge.

pjcozzi · 2015-06-22T11:47:31Z

@kainino0x also, remind me why we decided not to use the OBB for binning into multiple frustums? I want to keep some notes in case we decide to do it later.

kainino0x · 2015-06-22T14:47:11Z

@pjcozzi I just pushed a commit to the blog post which I had forgotten to push on Friday. Blog and PR should be ready.

I don't know that we made a specific decision not to use the OBB for binning into multiple frustums. It's possible that we just decided when it came up that I should keep working on other parts first.

kainino0x · 2015-06-22T14:56:44Z

Oh, I also forgot to push the heightmapped terrain commit to this repository on Friday. Please review that short commit. @pjcozzi

pjcozzi · 2015-06-23T22:33:09Z

@kainino0x can you merge in master?

This looks good. @kainino0x are you OK with us merging this so it can sit in master for a few days before the 1.11 release on July 1? Any additional OBB optimizations can come in a separate PR.

kainino0x · 2015-06-24T16:52:53Z

Sure. I'll sync with master when I get home today.

kainino0x · 2015-06-24T21:26:55Z

Done. I'm seeing a bunch of errors but they are also there on master. Plus the timeout ones that I see regularly.

mramato · 2015-06-24T22:42:00Z

The failures in master are a website issue. I'm putting in a fix now.

pjcozzi · 2015-06-25T12:29:52Z

Thanks @kainino0x. If we want to tweak this before the 1.11 release, we have until early next week.

Oriented bounding box & frustum culling optimizations

pjcozzi reviewed Jun 3, 2015
View reviewed changes

kainino0x force-pushed the oriented-bounding-box branch 2 times, most recently from c173ebb to de9094e Compare June 3, 2015 19:16

kainino0x reviewed Jun 3, 2015
View reviewed changes

kainino0x force-pushed the oriented-bounding-box branch 2 times, most recently from 09e1602 to 28c8c9d Compare June 3, 2015 20:19

kainino0x force-pushed the oriented-bounding-box branch from be25c6a to 7793970 Compare June 9, 2015 15:32

kainino0x changed the title ~~Oriented bounding box~~ Oriented bounding box & frustum culling optimizations Jun 9, 2015

kainino0x force-pushed the oriented-bounding-box branch from 7793970 to 8b8c5b8 Compare June 9, 2015 16:02

kainino0x force-pushed the oriented-bounding-box branch 2 times, most recently from 7b56349 to 0ed700e Compare June 9, 2015 19:37

kainino0x added 2 commits June 10, 2015 09:09

Replace .intersect(Cartesian4) with .intersectPlane(Plane)

dfebc12

This affects bounding volumes AxisAlignedBoundingBox and BoundingSphere.

Fix for ObjectOrientedBoundingBox.equals

2e81e95

Fix crash for debug drawing code

2f520c6

Add a comment about excluding skirts from OBBs

fb64614

kainino0x force-pushed the oriented-bounding-box branch from 53b46b4 to fb64614 Compare June 19, 2015 16:59

kainino0x added 3 commits June 19, 2015 13:55

Possible minor optimization for OBB.intersectPlane

0a712f6

Tile debug volume display: add comment and rename functions to clarif…

1cd0747

…y semantics

Add OBB performance note to CHANGES.md

85fc36c

Use bounding boxes for heightmap terrain (including no terrain).

d6a7389

Merge branch 'master' into oriented-bounding-box

37ae29a

pjcozzi added a commit that referenced this pull request Jun 25, 2015

Merge pull request #2782 from kainino0x/oriented-bounding-box

22f7dfc

Oriented bounding box & frustum culling optimizations

pjcozzi merged commit 22f7dfc into CesiumGS:master Jun 25, 2015

pjcozzi deleted the oriented-bounding-box branch June 25, 2015 12:29

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Oriented bounding box & frustum culling optimizations #2782

Oriented bounding box & frustum culling optimizations #2782

kainino0x commented Jun 3, 2015

pjcozzi Jun 3, 2015

kainino0x Jun 3, 2015

pjcozzi Jun 3, 2015

pjcozzi commented Jun 3, 2015

pjcozzi commented Jun 3, 2015

kainino0x Jun 3, 2015

pjcozzi commented Jun 3, 2015

kainino0x commented Jun 5, 2015

pjcozzi commented Jun 5, 2015

pjcozzi commented Jun 5, 2015

kainino0x commented Jun 9, 2015

pjcozzi commented Jun 9, 2015

kainino0x commented Jun 17, 2015

kainino0x commented Jun 18, 2015

kainino0x commented Jun 18, 2015

kring commented Jun 18, 2015

pjcozzi commented Jun 19, 2015

pjcozzi commented Jun 19, 2015

kainino0x commented Jun 19, 2015

kainino0x commented Jun 19, 2015

kainino0x commented Jun 19, 2015

pjcozzi commented Jun 19, 2015

pjcozzi commented Jun 21, 2015

pjcozzi commented Jun 22, 2015

kainino0x commented Jun 22, 2015

kainino0x commented Jun 22, 2015

pjcozzi commented Jun 23, 2015

kainino0x commented Jun 24, 2015

kainino0x commented Jun 24, 2015

mramato commented Jun 24, 2015

pjcozzi commented Jun 25, 2015

Oriented bounding box & frustum culling optimizations #2782

Oriented bounding box & frustum culling optimizations #2782

Conversation

kainino0x commented Jun 3, 2015

pjcozzi Jun 3, 2015

Choose a reason for hiding this comment

kainino0x Jun 3, 2015

Choose a reason for hiding this comment

pjcozzi Jun 3, 2015

Choose a reason for hiding this comment

pjcozzi commented Jun 3, 2015

pjcozzi commented Jun 3, 2015

kainino0x Jun 3, 2015

Choose a reason for hiding this comment

pjcozzi commented Jun 3, 2015

kainino0x commented Jun 5, 2015

pjcozzi commented Jun 5, 2015

pjcozzi commented Jun 5, 2015

kainino0x commented Jun 9, 2015

pjcozzi commented Jun 9, 2015

kainino0x commented Jun 17, 2015

kainino0x commented Jun 18, 2015

kainino0x commented Jun 18, 2015

kring commented Jun 18, 2015

pjcozzi commented Jun 19, 2015

pjcozzi commented Jun 19, 2015

kainino0x commented Jun 19, 2015

kainino0x commented Jun 19, 2015

kainino0x commented Jun 19, 2015

pjcozzi commented Jun 19, 2015

pjcozzi commented Jun 21, 2015

pjcozzi commented Jun 22, 2015

kainino0x commented Jun 22, 2015

kainino0x commented Jun 22, 2015

pjcozzi commented Jun 23, 2015

kainino0x commented Jun 24, 2015

kainino0x commented Jun 24, 2015

mramato commented Jun 24, 2015

pjcozzi commented Jun 25, 2015