3D Tiles - refine to visible children in replacement refinement #4287

lilleyse · 2016-09-06T21:12:57Z

Performance: optimize replacement refinement - instead of requesting all children when a tile needs to refine, request only the visible ones, and then allow refining a tile if all the required visible tiles are loaded (instead of all - including invisible - tiles). When the camera moves, and children that are not yet loaded are needed, either render the parent or render the available children and the parent with clipping.

When not all visible children are loaded, I go with a third approach - render the parent and available children together, which is slightly noticeable in some cases but doesn't hurt visual quality since the other visible children usually load in soon after.

Since the visibility check must now happen in the replacement refinement path, I save the result and don't recheck the visibility once the tile is popped from the stack. I modified the additive refinement path to do the same, store the planeMask now and use later.

I also ended up reverting #4280 because otherwise the bounding volume check will happen twice, first in the selectTiles loop, and again in selectTile. This only applies to tiles without content bounding volumes, which is many replacement tilesets. As long as contentsVisibility is called after visibility is already checked, this should be ok.

Some numbers:

Now that we don't need to request all children, and instead just the visible ones, there are some memory improvements. In real world tilesets I've noticed between 14% to 50% savings in the number of loaded tiles. These numbers below are for a generic replacement refinement tileset:

Before

View	Loaded	Rendered	Total tiles in tileset
Horizon	233	152	341
Straight Down	135	53	341

After

View	Loaded	Rendered	Total tiles in tileset
Horizon	210	152	341
Straight Down	86	53	341

This reverts commit 7d31334.

pjcozzi · 2016-09-06T23:52:15Z

I go with a third approach - render the parent and available children together, which is slightly noticeable in some cases but doesn't hurt visual quality since the other visible children usually load in soon after.

Can you please test this with several different datasets to get a feel for how bad it can be so we know how to prioritize clipping parents with their loaded children?

pjcozzi · 2016-09-06T23:54:30Z

As long as contentsVisibility is called after visibility is already checked, this should be ok.

Add a comment for this inside contentsVisibility and the reference doc (which is likely to be skipped when stepping debugging).

pjcozzi · 2016-09-06T23:56:37Z

86% savings

More like 14%, right?

Also, what is the Total column? It is always 341.

pjcozzi · 2016-09-06T23:58:27Z

Source/Scene/Cesium3DTile.js

@@ -294,6 +294,13 @@ define([
        this.replaced = false;

        /**
+         * The stored plane mask from the visibility check during tree traversal.
+         *
+         * @type {Boolean}


Isn't this CullingVolume?

pjcozzi · 2016-09-07T00:16:27Z

Specs/Scene/Cesium3DTilesetSpec.js

@@ -536,6 +514,42 @@ defineSuite([
        });
    });

+    it('replacement refinement - refines to visible ready children', function() {


With all the tileset traversal changes, are you sure this is the only new test we need? Are all cases covered?

Even though the test seems simple it does cover all the new traversal cases.

lilleyse · 2016-09-07T14:03:09Z

More like 14%, right?

Hah yeah, I updated the comment.

Also, what is the Total column? It is always 341.

Changed the name from "Total" to "Total tiles in tileset". It's always 341.

lilleyse · 2016-09-07T14:50:39Z

Can you please test this with several different datasets to get a feel for how bad it can be so we know how to prioritize clipping parents with their loaded children?

Some datasets are more noticeable than others, I think we will need to explore parent clipping to make this better.

lilleyse · 2016-09-07T15:28:14Z

Updated. I still need to add a flag to enable/disable this optimization.

pjcozzi · 2016-09-07T15:33:06Z

I still need to add a flag to enable/disable this optimization.

Will review and merge when this is ready.

lilleyse · 2016-09-07T18:34:09Z

Updated.

pjcozzi · 2016-09-07T18:46:18Z

Specs/Scene/Cesium3DTilesetSpec.js

@@ -536,6 +514,42 @@ defineSuite([
        });
    });

+    it('replacement refinement - refines to visible ready children', function() {
+        viewRootOnly();
+        return Cesium3DTilesTester.loadTileset(scene, tilesetUrl).then(function(tileset) {


Run this test twice, once with refineToVisible set to true and once with it set to false?

pjcozzi · 2016-09-07T19:11:14Z

Do you get these test failures with http://cesium-dev.s3-website-us-east-1.amazonaws.com/cesium/3d-tiles-refine-visible/Specs/SpecRunner.html?spec=Scene%2FCesium3DTileset

Scene/Cesium3DTileset does not select tiles when outside of view frustum
Failed: promise rejected: DeveloperError: parentPlaneMask is required.
DeveloperError@http://cesium-dev.s3-website-us-east-1.amazonaws.com/cesium/3d-tiles-refine-visible/Source/Core/DeveloperError.js:44:19

Scene/Cesium3DTileset replacement refinement - refines to visible ready children
Expected 5 to equal 2.
stack@http://cesium-dev.s3-website-us-east-1.amazonaws.com/cesium/3d-tiles-refine-visible/ThirdParty/jasmine-2.2.0/jasmine.js:1475:17

Scene/Cesium3DTileset tile visible event is raised
Failed: promise rejected: DeveloperError: parentPlaneMask is required.
DeveloperError@http://cesium-dev.s3-website-us-east-1.amazonaws.com/cesium/3d-tiles-refine-visible/Source/Core/DeveloperError.js:44:19

Scene/Cesium3DTileset tileUnload event is raised
Failed: promise rejected: DeveloperError: parentPlaneMask is required.
DeveloperError@http://cesium-dev.s3-website-us-east-1.amazonaws.com/cesium/3d-tiles-refine-visible/Source/Core/DeveloperError.js:44:19

pjcozzi · 2016-09-07T19:11:21Z

Just those comments.

lilleyse · 2016-09-07T19:14:05Z

Those test failures are fixed now, and a new test is added.

pjcozzi · 2016-09-07T21:25:27Z

Thanks @lilleyse! Huge improvement here.

JPhPons · 2017-03-01T14:29:35Z

Hi,
I observe a strange phenomenon on our 3D tiles of Marseille with a recent version of the 3d-tiles branch. Sometimes, tiles with very low resolution are displayed together with high resolution tiles. See the attached screenshot:

It should not be related to this PR, as I do not activate the refineToVisible option... Could it be an unwanted side-effect of this change, or another optimization in the replacement refinement?
Thank you in advance for your help,
Jean-Philippe

lilleyse · 2017-03-01T14:46:36Z

@austinEng can you verify if this is related to the child-union optimization?

lilleyse · 2017-03-01T20:11:37Z

@JPhPons if you pull the latest 3d-tiles branch and construct the tileset with cullWithChildrenBounds as false, do you see the same artifacts?

JPhPons · 2017-03-02T07:43:44Z

Hi Sean, Yes, I still have the same problem after proceeding as you suggested. JP From: Sean Lilley [mailto:notifications@github.com] Sent: mercredi 1 mars 2017 21:12 To: AnalyticalGraphicsInc/cesium <cesium@noreply.github.com> Cc: Jean-Philippe Pons <Jean-Philippe.Pons@bentley.com>; Mention <mention@noreply.github.com> Subject: Re: [AnalyticalGraphicsInc/cesium] 3D Tiles - refine to visible children in replacement refinement (#4287) @JPhPons<https://github.com/JPhPons> if you pull the latest 3d-tiles branch and construct the tileset with cullWithChildrenBounds as false, do you see the same artifacts? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#4287 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AY6Am9eXG_jZ6jnoDteugNDSwJjU7WRvks5rhdD-gaJpZM4J2Snf>.

lilleyse · 2017-03-02T15:05:42Z

Ok thanks for checking. Did you start to experience this problem recently? Is it occurring with other tilesets? It reminds me of issues we used to see with replacement refinement with empty tiles (like #3517).

Feel free to send me the tileset for testing.

pjcozzi · 2017-03-06T16:25:58Z

@JPhPons @lilleyse any update here?

JPhPons · 2017-03-31T12:41:42Z

Hi, Sorry for the late answer, I finally had time to look into it! I seems to be a side-effect of the pull request #3517, which solved only partially our problems with external tilesets. The problem has been there since then, it is a shame I didn’t notice it in my tests earlier… Please find attached a patch which fixes the problem. I am not confident enough to submit it before getting @lilleyse<https://github.com/lilleyse>‘s feedback. The patch can be tested in this demo: https://d3h9zulrmcj1j6.cloudfront.net/CesiumMeasurements/index.html JP De : Patrick Cozzi [mailto:notifications@github.com] Envoyé : lundi 6 mars 2017 17:26 À : AnalyticalGraphicsInc/cesium <cesium@noreply.github.com> Cc : Jean-Philippe Pons <Jean-Philippe.Pons@bentley.com>; Mention <mention@noreply.github.com> Objet : Re: [AnalyticalGraphicsInc/cesium] 3D Tiles - refine to visible children in replacement refinement (#4287) @JPhPons<https://github.com/JPhPons> @lilleyse<https://github.com/lilleyse> any update here? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#4287 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AY6AmyRInEZXQ5osFOQAPW34ROcIgYF6ks5rjDObgaJpZM4J2Snf>.

lilleyse · 2017-03-31T14:16:14Z

The attachment doesn't seem to go through with github replies, but you can send it to me directly or otherwise open a PR.

As a side note, the 3d-tiles traversal is about to get a large overhaul and likely fixes the issues here - #5128

lilleyse · 2017-03-31T15:45:13Z

Thanks for sending it over. The diff seems fine as an interim fix. Would you like me to open a PR?

JPhPons · 2017-03-31T16:10:10Z

I am still a newbie with github so that would be very helpful! Thanks, JP De : Sean Lilley [mailto:notifications@github.com] Envoyé : vendredi 31 mars 2017 17:45 À : AnalyticalGraphicsInc/cesium <cesium@noreply.github.com> Cc : Jean-Philippe Pons <Jean-Philippe.Pons@bentley.com>; Mention <mention@noreply.github.com> Objet : Re: [AnalyticalGraphicsInc/cesium] 3D Tiles - refine to visible children in replacement refinement (#4287) Thanks for sending it over. The diff seems fine as an interim fix. Would you like me to open a PR? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#4287 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AY6Am0xqqQdrMcVvYfDbOkZScDxp_fe6ks5rrR-OgaJpZM4J2Snf>.

lilleyse · 2017-03-31T16:17:46Z

Opened #5171

lilleyse added 6 commits September 6, 2016 15:24

Small fixes

d98c7cd

Only touch when out of core

a173ad0

Only refine to visible children for replacement refinement

fe77f16

Revert "contentsVisibility works for empty tiles"

daf6fb6

This reverts commit 7d31334.

Added test

df0564f

Revert other change

dd35d35

pjcozzi reviewed Sep 6, 2016
View reviewed changes

Whitespace

ed4fdf8

pjcozzi reviewed Sep 7, 2016
View reviewed changes

lilleyse added 2 commits September 7, 2016 10:56

Organization

261f0d1

Rename planeMask to visibilityPlaneMask

1763e76

lilleyse added 2 commits September 7, 2016 13:40

Remove comment

2b3f8e3

Add flag for refineToVisible

a57e2ca

Update comment

04ae86d

pjcozzi reviewed Sep 7, 2016
View reviewed changes

Add refineToVisible to 3D Tiles Sandcastle example

0a08889

Added test

9d5d56b

Whitespace

870dd86

pjcozzi merged commit e510c5b into 3d-tiles Sep 7, 2016

pjcozzi deleted the 3d-tiles-refine-visible branch September 7, 2016 21:25

pjcozzi mentioned this pull request Sep 7, 2016

3D Tiles Roadmap #3241

Open

pjcozzi mentioned this pull request Jan 11, 2017

optimization of replacement refinement selection/loading: only request visible tile children #3449

Closed

austinEng mentioned this pull request Mar 1, 2017

add flag to disable children union optimization #5058

Merged

lilleyse mentioned this pull request Mar 31, 2017

3D Tiles - External tileset fix #5171

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

3D Tiles - refine to visible children in replacement refinement #4287

3D Tiles - refine to visible children in replacement refinement #4287

lilleyse commented Sep 6, 2016 •

edited

pjcozzi commented Sep 6, 2016

pjcozzi commented Sep 6, 2016

pjcozzi commented Sep 6, 2016

pjcozzi Sep 6, 2016

pjcozzi Sep 7, 2016

lilleyse Sep 7, 2016

lilleyse commented Sep 7, 2016 •

edited

lilleyse commented Sep 7, 2016

lilleyse commented Sep 7, 2016

pjcozzi commented Sep 7, 2016

lilleyse commented Sep 7, 2016

pjcozzi Sep 7, 2016

pjcozzi commented Sep 7, 2016

pjcozzi commented Sep 7, 2016

lilleyse commented Sep 7, 2016

pjcozzi commented Sep 7, 2016

JPhPons commented Mar 1, 2017

lilleyse commented Mar 1, 2017

lilleyse commented Mar 1, 2017

JPhPons commented Mar 2, 2017 via email

lilleyse commented Mar 2, 2017

pjcozzi commented Mar 6, 2017

JPhPons commented Mar 31, 2017 via email

lilleyse commented Mar 31, 2017

lilleyse commented Mar 31, 2017

JPhPons commented Mar 31, 2017 via email

lilleyse commented Mar 31, 2017

3D Tiles - refine to visible children in replacement refinement #4287

3D Tiles - refine to visible children in replacement refinement #4287

Conversation

lilleyse commented Sep 6, 2016 • edited

pjcozzi commented Sep 6, 2016

pjcozzi commented Sep 6, 2016

pjcozzi commented Sep 6, 2016

pjcozzi Sep 6, 2016

Choose a reason for hiding this comment

pjcozzi Sep 7, 2016

Choose a reason for hiding this comment

lilleyse Sep 7, 2016

Choose a reason for hiding this comment

lilleyse commented Sep 7, 2016 • edited

lilleyse commented Sep 7, 2016

lilleyse commented Sep 7, 2016

pjcozzi commented Sep 7, 2016

lilleyse commented Sep 7, 2016

pjcozzi Sep 7, 2016

Choose a reason for hiding this comment

pjcozzi commented Sep 7, 2016

pjcozzi commented Sep 7, 2016

lilleyse commented Sep 7, 2016

pjcozzi commented Sep 7, 2016

JPhPons commented Mar 1, 2017

lilleyse commented Mar 1, 2017

lilleyse commented Mar 1, 2017

JPhPons commented Mar 2, 2017 via email

lilleyse commented Mar 2, 2017

pjcozzi commented Mar 6, 2017

JPhPons commented Mar 31, 2017 via email

lilleyse commented Mar 31, 2017

lilleyse commented Mar 31, 2017

JPhPons commented Mar 31, 2017 via email

lilleyse commented Mar 31, 2017

lilleyse commented Sep 6, 2016 •

edited

lilleyse commented Sep 7, 2016 •

edited