HyVGGT-VO: Hybrid Dense Visual Odometry With Classical VO and Feed-Forward Model

arXiv:2604.02107v2 Announce Type: replace Abstract: Dense visual odometry (VO), which provides pose estimation and dense 3D reconstruction, serves as the cornerstone for applications ranging from robotics to augmented reality. Recently, feed-forward models have demonstrated remarkable capabilities in dense mapping. However, when these models are used in dense visual SLAM systems, their keyframe-based processing often leads to sparse pose outputs, limiting continuous per-frame pose estimation. In contrast, traditional sparse methods provide high-frequency pose outputs but lack the capability for dense reconstruction. To address these limitations, we propose HyVGGT-VO, a novel framework that combines the computational efficiency of sparse VO with the dense reconstruction capabilities of feed-forward model. To the best of our knowledge, this is the first work to tightly couple a traditional VO frontend with VGGT. Specifically, we design an adaptive hybrid tracking frontend that dynamically switches between traditional optical flow and the VGGT model to ensure robustness. Furthermore, we introduce a hierarchical optimization framework that jointly refines VO poses and the scale of VGGT predictions to ensure global scale consistency. Under the odometry setting, compared with VGGT-SLAM 2.0, HyVGGT-VO reduces the average ATE by approximately 91% on EuRoC MAV and 20% on TUM RGB-D, and decreases the average relative translation and rotation errors by 46% and 92% on KITTI, respectively, while achieving 16.1 Hz pose output and 3.2 Hz dense-mapping updates with 6.5 GB of memory on an 8 GB laptop GPU.Our code is available at https://geneta2580.github.io/HyVGGT-VO.io.