`ModelExperimental` Sandcastle Testing

[ptrgags]

ModelExperimental Sandcastles

As part of https://github.com/CesiumGS/cesium/issues/10346, I did a sweep through the old 3D Tiles and Model Sandcastles to see what happens when I set enableModelExperimental = true or use ModelExperimental.fromGltf instead of Model.fromGltf.

EDITED: 07/08/2022

3D Tiles Sandcastles

Working Sandcastles:

• 3D Tiles Adjust Height
• 3D Tiles Batch Table Hierarchy
• 3D Tiles Compare
• 3D Tiles Feature Picking
• 3D Tiles Feature Styling
• 3D Tiles Formats
• 3D Tiles Inspector
• 3D Tiles Interior
• 3D Tiles Photogrammetry
• 3D Tiles Point Cloud Classification
• 3D Tiles Point Cloud Shading
• 3D Tiles Point Cloud
• 3D Tiles Terrain Classification
• Clamp to 3D Tiles
• Sample Height from 3D Tiles
• Development/3D Tiles Split

Broken Sandcastles:

• 3D Tiles BIM
  • Shader compilation error. Expected, this is glTF 1.0
• ~3D Tiles Batch Table Hierarchy~ -- fixed in #10525
  • "Color all doors" doesn't work, as batchTable.isExactClass is not a function
  • "Color all features derived from door" doesn't work, as batchTable.isClass is not a function
• 3D Tiles Clipping Planes
  • Fails due to the glTF 1.0 BIM model
• ~3D Tiles Compare~ fixed in #10506
  • OSM Buildings have incorrect shading (see #10338)
• ~3D Tiles Feature Styling~ fixed in #10514
  • OSM Buildings have incorrect shading
  • Color by Distance to Selected Location crashes shader -- fixed by #10514
• ~3D Tiles Formats~ fixed in #10524
  • Wireframe mode will need enableDebugWireframe: true
  • Composite tileset is repeated for some reason
• ~3D Tiles Inspector~ fixed in #10526
  • Wireframe mode will need enableDebugWireframe: true
  • Statistics are not available (not yet implemented)
• ~3D Tiles Interior~ fixed in #10524
  • Some parts of the model (most notably the floor!) are not loading
• 3D Tiles Photogrammetry Classification
  • The classification isn't showing up
• 3D Tiles Point Cloud Styling
  • Many of the styles are failing due to an undefined property value
• Polylines on 3D Tiles
  • Fails due to glTF 1.0 BIM model
• ~Development/3D Tiles Split~ fixed in #10524
  • still using .otherwise()
  • Again, composite model renders multiple times

Other 3D Tiles Sandcastles

These Sandcastles were found by searching "Cesium3DTileset" and "osmBuildings" in the Sandcastle searchbar

Working Sandcastles:

• Classification Types
• Classification
• Cesium OSM Buildings
• Clouds
• FXAA
• Fog Post Process
• Scene Rendering Performance
• Terrain Clipping Planes
• development/3D Tiles Performance Testing
• development/Many Clipping Planes
• development/Pick From Ray
• development/Picking
• development/Shadows
• CZML 3D Tiles

Broken Sandcastles:

• Ambient Occlusion
  • Fails due to glTF 1.0 BIM model
• MSAA
  • Fails due to glTF 1.0 BIM model
• Montreal Point Cloud
  • Point cloud styling not showing up (GPU styling not implemented). See also #9944
• ~Cesium OSM Buildings~ - fixed in #10506
  • Incorrect lighting of OSM Buildings
  • No outlines
• ~Clouds~ fixed in #10506
  • Incorrect Lighting of OSM Buildings

Model Sandcastles

For these ones, I replaced Model.fromGltf with ModelExperimental.fromGltf

Working Sandcastles:

• Manually Controlled Animation
• development/Display Conditions
• development/Multiple Shadows

Broken Sandcastles:

• HeadingPitchRoll
  • Plane is frozen in place.
  • The camera doesn't zoom to the model correctly
• Image-Based Lighting
  • Doesn't seem to load the environment map. The sandcastle might need more adjustment to the new IBL API
  • The camera doesn't zoom in correctly
• LocalToFixedFrame
  • The camera zooming is incorrect
• development/3D Models Articulations
  • 404 on asset from assets.agi.com
• development/3D Models Instancing
  • ModelInstanceCollection... how will that work with ModelExperimental?
• development/3D Models Node Explorer
  • nodesByName is undefined
• development/3D Models
  • Again, the camera zooming doesn't work properly

[ptrgags]

Today I looked into why the glTF 1.0 BIM model crashes in ModelExperimental.

There's one shader that declares the normals as attribute vec2 a_n; and does the oct decoding in the technique's vertex shader.

Trying to allow a special case for this in ModelExperimental will add considerable complexity, so I wouldn't recommend we do that. Better to update the asset to use glTF 2.0

Some thoughts about how to update the model:

• Probably the easiest thing would be to decode the oct-encoded normals an just use the standard NORMAL attribute.
• The techniques would have to be converted to regular glTF materials.
• If the shaders in the techniques are still desired, they could be manually converted to a CustomShader. Though you'd need a _FEATURE_ID attribute to determine which shader to apply.

Unfortunately this upgrade process would have to be done on a case-by-case basis since glTF techniques are very open-ended (depends highly on the GLSL shader used)

[j9liu]

Came up in #10388: Model keeps its boundingSphere in local space, whereas ModelExperimental returns the bounding sphere in world space (i.e. with modelMatrix applied). The ModelExperimental method is better and more consistent with Cesium3DTileset, we just need to update the Model Sandcastles with this in mind and document it CHANGES.md