Geometry Reconstruction · Hand Reconstruction

Multimodal scene geometry reconstruction fuses multi-sensor data (e.g., RGB, depth, and IMU) to model hands, objects, and complex environments with high precision while jointly estimating global sensor poses. This provides reliable perception foundations and fine-grained annotations for embodied intelligence. Our team has achieved high-precision hand reconstruction under RGB-D settings; upcoming work will incorporate scene priors to further improve hand pose estimation and extend to category-level object pose estimation.