More accurate rendering of models in 2D

[lilleyse]

The larger the model, the less accurate it will be represented in 2D since it does not use the model's latitude in determining a stretching factor. Maybe just a scale will work, or maybe its harder than that since certain areas of the model vary by their latitude. @bagnell you probably have a better grasp of the details here.

I remember this being a known limitation when 2D model support was added. It's something that's worth more investigation considering many of our models will actually be large city sections.

From this forum post: https://groups.google.com/forum/#!topic/cesium-dev/jBmwFWHgVDo

3D:

Columbus View:

[bagnell]

Yes, this was a known limitation when 2D support was added for models. For large models, we would need to go through the positions and project them to 2D. The farther the positions from the origin in model coordinates, the more inaccurate it will be in 2D.

[pjcozzi]

Maybe with emulating double-precision in the vertex shader (and with transform-feedback in WebGL 2.0, #797) the re-projection could be done on the GPU; it is not affine so it can't be burned into the model material.

Otherwise, I think we would have to use extra memory to store 3D and 2D positions.

[jadelhamilton]

Thanks I will look in to it and find something

…

On Thu, Mar 16, 2017 at 6:41 PM Sean Lilley ***@***.***> wrote: The larger the model, the less accurate it will be represented in 2D since it does not use the model's latitude in determining a stretching factor. Maybe just a scale will work, or maybe its harder than that since certain areas of the model vary by their latitude. @bagnell <https://github.com/bagnell> you probably have a better grasp of the details here. I remember this being a known limitation when 2D model support was added. It's something that's worth more investigation considering many of our models will actually be large city sections. From this forum post: https://groups.google.com/forum/#!topic/cesium-dev/jBmwFWHgVDo 3D: [image: 3d] <https://cloud.githubusercontent.com/assets/915398/24023068/51ff543c-0a80-11e7-8433-20aa085f4a85.PNG> Columbus View: [image: auto generated inline image 1] <https://cloud.githubusercontent.com/assets/915398/24023071/573c81c2-0a80-11e7-8445-ff3b6885310b.png> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5112>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJG2kQ69Wfpb-r7-24NuZZi5b_TY7dv4ks5rmcipgaJpZM4MgA4j> .

[lilleyse]

Possibly affecting this photogrammetry tileset too, from https://groups.google.com/forum/#!topic/cesium-dev/KxpSBtfKetk.

However not noticeable for the AGI photogrammetry tileset.

[lilleyse]

This came up on the forum: https://groups.google.com/forum/#!topic/cesium-dev/nq5SymE_PPg

[lilleyse]

@j9liu can this be closed now that #10384 is merged?

[j9liu]

I'll close this, yes