Model loading performance problem of Forge AR/VR toolkit

Question

During the summer of 2018, when I tried the Forge AR/VR Toolkit, it was found that loading a model into Unity takes much longer time(several hours) than doing the same with the Forge web API(a few seconds). At that moment the AR/VR toolkit was just an preview version. Now it's almost the end of 2019 and I can see that there are severval updates on AR/VR toolkit, and with the latest version (ForgeARKit-update-6-2018.1.unitypackage), the loading performance is improved a lot). But it still takes a long time, depending on the size of the original model. eg For a model which is about 100m, it takes more than 1 hour to load it into Unity. I am wondering whether we can expect Forge to support model optimization/zipping in future release, the expectation is simple: being able to load model in Unity/AR/VR as fast as loading in web. Frankly speaking we are considering ultilizing third party service such as 'Tridify' instead of Forge, if it we cannot expect what what we expected.

In addition, it seems both the 'AR/VR toolkit Unity package' and the 'scene preparation sample' hasn't been update for serveral months. Is there any change to the roadmap? On the other side, from the demo of 'Autodesk reflect' which is released recently, it looks that it's already possible to load model in Unity/AR/VR efficiently. But isn't that something what Forge AR/VR toolkit should provide?? Please clarify, just a bit confused.

Answer 1

The Forge Toolkit v1 is slow based on the file content. For example, if your CAD file contains thousands of little objects, it would be terribly slow. This is something we are addressing in version v2 which will come out in January. You should expect to see an improvement on the speed to match the Forge Viewer speed with that version. It will also have major enhancements such as spatial and property queries, instancing and deduplication, and a connection to BIM360 and OTG. I also want to repeat that the toolkit v1 was designed for partial loading and not full model loading, but we saw that people use it for full model loading scenarios, this is why we are addressing the issue in the v2 version.