3D Tiles - Point Cloud color compression #4315
Conversation
+1 for staying with the simple approach, especially given they are the same size and that RGB565 is a common WebGL format (texture, not attribute AFAIK; regardless, it is still familiar). Did you just eyeball the difference or did you diff them or find any reading on perception to compare the two? |
|
@hobu @connormanning RGB565 is an optional lossy compressed per-point color format that you may want to check out to store colors in 16 bits. |
| @@ -520,6 +532,16 @@ define([ | |||
| if (hasColors) { | |||
| if (isTranslucent) { | |||
| vs += 'attribute vec4 a_color; \n'; | |||
| } else if (isYCoCg || isRGB565) { | |||
| vs += 'attribute float a_color; \n' + | |||
| 'const float SHIFT_RIGHT_11 = 1.0 / 2048.0; \n' + | |||
There was a problem hiding this comment.
I feel like passing on this for now.
There was a problem hiding this comment.
OK, added to the roadmap. Good beginner low-hanging fruit.
|
Just those comments. Let me know when this is ready with tests for review. Can you please also open a PR for the spec? |
Yeah, just ran my demo and compared the results visually. They both have banding, just in different ways. |
lilleyse
force-pushed
the
color-compression
branch
from
e5f61d8
to
2cc0901
Compare
lilleyse
force-pushed
the
color-compression
branch
from
2cc0901
to
d40a3b9
Compare
|
Removed ycocg and added a test. This is ready now. I'll update the spec. |
I was hoping for something more scientific, but given that RGB565 is widely used in GL, I think it is appropriate. Other options can be added later, of course. |
|
Spec update: CesiumGS/3d-tiles#130 |
For #3241 and CesiumGS/3d-tiles#22 (comment)
I tried out two different ways of compressing point cloud colors
Results:
http://cesium-dev.s3-website-us-east-1.amazonaws.com/cesium/color-compression/Apps/Sandcastle/?src=Hello%20World.html&label=Showcases&gist=29c700b58600b669fbbf9e30aeeac914
Overall I don't think YCoCg offers much benefit and may not be suited for this purpose. The RGB565 and YCoCg results are pretty similar, and conceptually RGB565 is simpler.
Note: once people have a chance to check out the demo and review the code I'm going to remove the Temp folder and add the point clouds to the main folder, as well as write a simple test. I'll squash the changes so we don't have the large point clouds still in the repo afterwards.