Using only browser technology (ARCore or similar apps that interface with the browser do not count) and without relying on markers I want to be able to determine how big a 3d model should be in order to be true to size relative to the objects in a camera feed. Ideally without the user providing any additional information like how far the detected plane is from their mobile device. From what I can tell there is no good solution at the moment of writing (WebXR uses ARCore under the hood so that's a no go). Is this something browsers are working to solve? What discussions can I look at on this topic? I feel like this is not a new problem but I don't know where to go for the latest information. Will browser-based size accurate AR without markers/user input ever be possible?
I heard some rumors about neural nets being employed to provide accurate measuring of camera space. I don't know where those came from so if anyone has any more details on that this would be a good place to put it.
Please comment if you are you aware of any forum discussions/feature requests about augmenting browsers to be able to provide the information to make accurate estimates of surface distance, room size, etc.... I want to get involved and help drive that movement.
I pray this question finds it's way to the people who know stuff. Thank you to all the people answering questions out there. You guys are awesome!
The best option I've seen is 8th Wall's True Scale feature within their world-tracking library. This is not an open-source solution, it is paid but it is the best in the industry for their SLAM tracking.
From my understanding the 8th Wall true scale runs completely in browser and uses some AI algorithms to detect the environment and appropriately scale in the 3D content in the scene.