Details for AnchorFlow VIO
Submission data
| Full name | AnchorFlow Visual-Inertial Odometry |
| Description | Basalt-based monocular visual-inertial odometry using the left grayscale camera and right IMU. Uses fixed-template optical-flow tracking, factory IMU rectification and a uniform configuration across
sequences. No stereo measurements, loop closure or offline global refinement. Original AnchorFlow contributions are MIT licensed; Basalt and other dependencies retain their respective licenses and acknowledgments. Reported throughput is a historical aggregate native offline rate: 265,694 timestamps in 3,139.15 seconds, including observation dumping but excluding input preparation and evaluation. CPU-affinity conditions varied; this is not a controlled deployment or main-test speed measurement. |
| Parameters | cam0 + imu0; fixed temporal templates; feature detection grid 50 px; outlier threshold 3 px; one-second initialization window; two estimator threads. Factory IMU rectification with 4.2 ms subtracted from IMU timestamps. Identical frozen configs/mono.json and configs/basalt.json across all sequences. |
| Programming language(s) | C++ and Python |
| Hardware | Intel Core i7-10700, 16 GB RAM, CPU only, two estimator threads |
| Throughput | 84.64 FPS |
| Website | https://github.com/sudhirpratapyadav/anchorflow-vio |
| Source code or download URL | https://github.com/sudhirpratapyadav/anchorflow-vio |
| Latest upload | 19 Oct 2026 |