LingBot-Depth overview
LingBot-Depth converts incomplete or noisy sensor depth into metric depth measurements and 3D points. It uses an RGB image, raw depth, and camera intrinsics as inputs.
Released models
| Checkpoint | Use |
|---|---|
| LingBot-Depth-v0.5 | Recommended checkpoint for general depth refinement and completion |
| LingBot-Depth-v0.1 | Earlier general depth refinement checkpoint |
| LingBot-Depth-DC | Checkpoint optimized for sparse depth completion |
The training data contains 3 million RGB-D samples from real scenes, robot manipulation data, and simulation.
Use Installation and Quick Start to run the included example. Use Data, applications, and model details for input and output formats, inference parameters, training data, and downstream examples.