We are looking into the migration of a product from Google Earth to Cesium. Part of the product involves moving a ship’s model in real time.
I have set up a small trial to move the ship at the clock ticks. If I use the Cesium_Ground.gltf model from the sandbox, the animation is very smooth; the frame rate is 40-50 FPS (according to the Cesium inspector).
I have also converted an existing ship model from .dae to .gltf with the offline conversion tool, the model is almost 9MB (vs. about 1MB for Cesium_Ground.gltf).
If I use this model in the same trial, the rate drops to 4FPS and the animation is not very smooth (i.e. not good enough for our purpose).
Can this be caused by the larger model, or is there something wrong with the model itself? (Unfortunately I can’t publish the model here).
Or are there larger .gltf models available somewhere that I could try out.
Most likely the model itself has a lot of separate nodes/meshes/primitives. In an extreme case, we have run into COLLADA models exported from SketchUp that had one triangle per node!
You can probably import the model into an authoring tool and then export with “combined nodes/meshes/primitives.”
Thank you for your reply. Our model has lots more meshes/primitives than the sandbox model (and this is not even our most detailed model…). The model was made in SketchUp and we will need to investigate if the number of meshes/primitives can be reduced, that will take some more time.
I was wondering if the planned KML support in release 1.7 (or later) could help us. Although I am not sure if support for models is planned for release 1.7 .
After all, the model(s) that I converted to .gltf are from KMZ files (created in SketchUp) that worked fine in Google Earth. So if the KML implementation will be different from the .gltf, performance may be better.
Second question; does the video card have an impact on the performance, we’re looking at a geforce gtx 550.
A GeForce GTX 550 will run Cesium well. Generally, any GeForce card from the past few years is OK. Most Cesium applications are actually CPU-bound, but the GPU can be the limiting factor in some cases like very high-resolution displays and large amounts of static geometry.
For KML, we will convert COLLADA to glTF on the fly with a web service similar to the online tool.