Method name | Active Search v1.1 w/ pose prior, image triplets |
Dataset | robotcar |
Submission date | April 10, 2020, 6:25 p.m. |
Last modified | April 10, 2020, 6:28 p.m. |
Publication url | https://ieeexplore.ieee.org/abstract/document/7572201 |
Code url | https://github.com/tsattler/vps |
Other info | Re-implemented RANSAC part (based on Lebeda et al., BMVC 2012), all poses are now evaluated for the benchmark (previously, only poses with >=12 inliers were considered), uses a pose prior: matches each query image only against the relevant sub-model, uses image triplets (left, rear, right images taken at the same time) for pose estimation |