Hacker News Reader

Comment by k__ | Hacker News Reader

Isn't the main selling point of the current AR iteration, that these markers aren't needed anymore?

There are still good reasons for markers, like syncing two different devices to the same augmented space.

Does anyone know if Apple's ARKit has a solution for this?

That's what I was hoping to see from the game demo during the keynote, but they didn't allude to it. That's the next key element we're going to need.

not yet. right now ARKit feature points are not 'persistent'. Meaning, they are only really used to help track 3d space frame over frame (with the internal gyro). Getting a 3d map that can be shared with multiple people is another layer of complexity not yet integrated into ARKit

No, the toy AR in smartphones doesn't really know depth or scale to the accuracy of a known marker. They just make assumptions and do basic plane-finding to avoid collisions. You don't have to know depth to make a hot dog sit on a plane, you just have to know it's a plane and how many pixels tall you want the hot dog.

It will be a long way before markers are obsolete. My guess is about 5 years.

Have you tried the "toy AR" in new Apple devices running iOS 11? it's quite impressive how accurately it's able to track most environments. Sure, having a known marker makes that process a lot easier, so you can avoid the issue entirely even where ARKit has problems (a large featureless surface, for example)...

But even with the current limitations, it does feel like the next evolution of AR -- one you can just start using with no setup. Printing a marker is not something a casual person, just interested in checking out a new app, would do (assuming they have access to a printer, or a second device, in the first place). I think that it's a different ball game, and agree with OP that it's the main point of the new generation.

That's not to say anything negative about this project specifically -- it's awesome, and really cool to see something like this perform well on the web in general, and mobile web in particular.

Yes, I develop monocular slam for a stealth startup, so I have directly dealt with the requirements for markerless AR. It's a well known fact in SLAM that a physical measurement is necessary to determine scale. IMUs get close, depth sensors are better but have rough resolution, seeing a known object (like a marker) in the space is best.

Because the toy AR displays on a small 2D screen, there's a number of shortcuts that can go mostly unnoticed.

Reply on news.ycombinator.com