1DSfM-derived rotation averaging benchmark
收藏资源简介:
本研究基于1DSfM数据集构建了一个新的旋转平均基准,旨在提供一个更可靠和全面的评估工具,用于评估旋转平均研究。该基准通过使用COLMAP库重新生成每个测试数据的Epipolar-geometry Graph(EG),并结合COLMAP和Bundler的交叉检查结果作为更可靠的地面实况源。此外,该基准还引入了新的评估任务和指标,不仅关注旋转估计精度,还考虑了估计旋转对下游位置估计任务的影响,从而全面评估不同旋转平均方法的性能。
This study develops a novel rotation averaging benchmark based on the 1DSfM dataset, which aims to provide a more reliable and comprehensive evaluation tool for rotation averaging research. This benchmark regenerates the Epipolar-geometry Graph (EG) for each test dataset using the COLMAP library, and leverages the cross-check results from COLMAP and Bundler as a more credible ground truth source. Furthermore, this benchmark introduces new evaluation tasks and metrics that not only concentrate on the accuracy of rotation estimation, but also take into account the influence of estimated rotations on downstream pose estimation tasks, thus comprehensively evaluating the performance of various rotation averaging methods.

- 1Incremental Rotation Averaging Revisited and More: A New Rotation Averaging Benchmark中国科学院自动化研究所 · 2024年



