WEBVTT

0:00:00,160 --> 0:00:05,920
We present AdjustAR, a system for runtime 
AI-driven adjustment of site-specific AR content.

0:00:05,920 --> 0:00:10,640
Authoring is often done with pre-captured 
3D models, which can become outdated.

0:00:10,640 --> 0:00:14,400
With AdjustAR, authors place 
content in these models as usual.

0:00:14,400 --> 0:00:19,040
At runtime, the system captures the live scene 
and compares it to the authored scene using a

0:00:19,040 --> 0:00:22,720
multimodal large language model.
When misalignments are detected,

0:00:22,720 --> 0:00:28,640
it proposes corrected placements using depth and 
camera data, which are then shown to the end-user.