Hi, first of all, thank you for the great project. I've been testing LingBot-Map for large indoor mapping scenarios and the reconstruction quality is quite impressive.
I noticed that in long trajectories, especially when returning to a previously visited area, there can be noticeable accumulated drift and the final map may not align well with the starting region. From my understanding, the current system seems to focus on local pose estimation and dense reconstruction, but I couldn't find information about loop closure detection or global pose graph optimization.
Do you have any plans to support loop closure, pose graph optimization (e.g. GTSAM/g2o), or other forms of global consistency optimization in future releases?
My use case is offline mapping, so I do not require real-time performance and would be happy to trade computation time for better global consistency.
Also, is there a recommended way to externally refine the predicted camera poses and then rebuild the map from the optimized poses? LingBot-Map already provides camera poses, depth maps, and dense world points, so it seems feasible to integrate an external backend optimizer.
Thanks for your work and looking forward to your thoughts.
你好，首先感谢你们开源这个项目。我最近一直在使用 LingBot-Map 进行大规模室内场景建图，整体重建效果给我留下了非常深刻的印象。
不过我发现，在较长轨迹的场景中，特别是最终回到曾经访问过的区域时，会出现明显的累计漂移，导致地图首尾无法完全对齐。从目前的实现来看，我理解项目主要关注局部位姿估计和稠密重建，但我没有找到关于回环检测（Loop Closure）或全局位姿图优化（Pose Graph Optimization）的相关说明。
想请问未来是否有计划支持：
回环检测（Loop Closure）
Pose Graph Optimization（例如 GTSAM 或 g2o）
其他用于提升全局一致性的优化方法
我的使用场景主要是离线建图，因此并不要求实时性能，更希望通过增加计算时间来获得更好的全局一致性。
另外，目前是否有官方推荐的方式，对 LingBot-Map 输出的相机位姿进行外部优化，然后基于优化后的 Pose 重新生成地图？因为项目已经能够输出相机位姿、深度图以及稠密世界坐标，看起来似乎具备与外部后端优化器集成的条件。
感谢你们的工作，期待你们的回复。
This is my setting : --use_sdpa --debug_wp --mode windowed --window_size 300 --overlap_keyframes 32 --keyframe_interval 1
Hi, first of all, thank you for the great project. I've been testing LingBot-Map for large indoor mapping scenarios and the reconstruction quality is quite impressive.
I noticed that in long trajectories, especially when returning to a previously visited area, there can be noticeable accumulated drift and the final map may not align well with the starting region. From my understanding, the current system seems to focus on local pose estimation and dense reconstruction, but I couldn't find information about loop closure detection or global pose graph optimization.
Do you have any plans to support loop closure, pose graph optimization (e.g. GTSAM/g2o), or other forms of global consistency optimization in future releases?
My use case is offline mapping, so I do not require real-time performance and would be happy to trade computation time for better global consistency.
Also, is there a recommended way to externally refine the predicted camera poses and then rebuild the map from the optimized poses? LingBot-Map already provides camera poses, depth maps, and dense world points, so it seems feasible to integrate an external backend optimizer.
Thanks for your work and looking forward to your thoughts.
你好，首先感谢你们开源这个项目。我最近一直在使用 LingBot-Map 进行大规模室内场景建图，整体重建效果给我留下了非常深刻的印象。
不过我发现，在较长轨迹的场景中，特别是最终回到曾经访问过的区域时，会出现明显的累计漂移，导致地图首尾无法完全对齐。从目前的实现来看，我理解项目主要关注局部位姿估计和稠密重建，但我没有找到关于回环检测（Loop Closure）或全局位姿图优化（Pose Graph Optimization）的相关说明。
想请问未来是否有计划支持：
回环检测（Loop Closure）
Pose Graph Optimization（例如 GTSAM 或 g2o）
其他用于提升全局一致性的优化方法
我的使用场景主要是离线建图，因此并不要求实时性能，更希望通过增加计算时间来获得更好的全局一致性。
另外，目前是否有官方推荐的方式，对 LingBot-Map 输出的相机位姿进行外部优化，然后基于优化后的 Pose 重新生成地图？因为项目已经能够输出相机位姿、深度图以及稠密世界坐标，看起来似乎具备与外部后端优化器集成的条件。
感谢你们的工作，期待你们的回复。
This is my setting :
--use_sdpa --debug_wp --mode windowed --window_size 300 --overlap_keyframes 32 --keyframe_interval 1