Con-DSO: Learning Short-Horizon Consistency Priors for RGB-D Direct Sparse Odometry

Name: Con-DSO: Learning Short-Horizon Consistency Priors for RGB-D Direct Sparse Odometry
Uploaded: 2026-10-07T22:54:32.861097+00:00
Duration: 512 s
Description: Con-DSO improves visual odometry by learning short-horizon consistency uncertainty from adjacent RGB-D frames. It predicts photometric and depthgeometric errors to create a host-side quality prior for keyframe tracking. This allows the system to selectively suppress unreliable observations caused by dynamic objects or poor illumination, leading to more robust pose estimation.

Con-DSO improves visual odometry by learning short-horizon consistency uncertainty from adjacent RGB-D frames. It predicts photometric and depthgeometric errors to create a host-side quality prior for keyframe tracking. This allows the system to selectively suppress unreliable observations caused by dynamic objects or poor illumination, leading to more robust pose estimation.

Open the full episode