3 ms·
I doubt it - the cameras are not at eye location. I expect they employ a similar solution to the Quest, which has no front-facing cameras at all - it builds a 3
by dTal 3y ago
I doubt it - the cameras are not at eye location. I expect they employ a similar solution to the Quest, which has no front-facing cameras at all - it builds a 3d model of the scene on the fly using photogrammetry on the corner cameras (if Apple do this they'd avail themselves of the built in LIDAR) and folds it into the rest of the rendering stack. This has theoretically better latency for the important part (head movement) than piping cameras directly into the displays, because it's perfectly synchronized with the display refresh without having to wait for a camera to complete a frame. I expect Apple does something similar.
- zaptrem 3y agoThis isn't what Quest does. Photogrammetry isn't that good (nor that fast). They just warp the images from the down/outward-facing cameras.
- dTal 3y agoWell sort of. It's a depth-accurate warp, fed from the SLAM system. This is functionally equivalent to texturing a depth map with the cameras. It has to be this way because there's no other way to simulate having a camera in a different spot than it really is. If they were simple 2d warps, parallax, occlusion, and stereopsis would be all wrong, and you'd feel violently ill any time you moved your head. You can see the system break down if you bring your hand too close to the headset. You're certainly not getting raw passthrough images. I'm not intimately familiar with the precise technical details of the Quest system, but I do know that real time photogrammetry even more sophisticated than depth maps can be achieved on mobile chipsets - google LSD-SLAM.
- zaptrem 3y agoThis makes a lot more sense, thanks for elaborating.