A phone's motion sensor fixes the scale of 3D vision models that otherwise can't tell true size from apparent size.

Computer models can figure out 3D scenes and camera angles from regular photos, but they struggle with real-world measurements. Because single-camera pictures lack depth scale, these systems cannot tell real distances accurately. Researchers built VI3 to fix this using inertial measurement unitsinertial measurement unitsSensors in devices like phones that measure motion and orientation without using cameras., or IMUs. IMUs are common sensors that track physical motion. VI3 takes motion data from an IMU, processes it into a metric motion guide, and uses that guide to calculate the true scale of the model's 3D output. It adapts to different model designs. Tests on synthetic and real aerial data show VI3 recovers real-world scale without needing ground-truth training data, keeps shapes consistent, refines good motion, and provides a strong guide when motion data is weak.

Why it matters

For anyone building apps with 3D mapping, like augmented reality or drone navigation, scale errors cause objects to appear the wrong size or position. VI3 lets these systems use cheap motion sensors already in devices to get accurate measurements without extra calibration.

Questions answered