kth-rpl/dynamicmap_benchmark

The First Dynamic Map Removal Benchmark: open-sourced all the methods and datasets in papers. Continous updating.

dynamicmap_benchmark's Issues

请问怎么生成逐帧的清理结果

你好 我发现removert代码只提供了全局地图的清理结果pcd文件 请问怎么生成每一帧 即每一个pcd文件的清理结果？

The octomapfg method yields the wrong results.

Hi, it's a nice work, I like it.

When I run your open source code(octomapfg method), I get the following result, the point cloud is all white, I don't get the same result as removert, what is the reason?

Wrong result

Correct result

问题交流！

Hello 张博，
我现在其实在做一些点云的数据标注工作，目前想法是参考semantic kitti，去融合多帧的点云，有一个比较麻烦的问题就是动态障碍物的去除上。但是由于场景中的动态障碍物比较少（比如在停车场的场景），可能动态物体只占1%的比例，比如行驶的车、人等，但很少。目前是用了一个MOT的方法去做的动态判断，把多帧的检测结果关联起来，通过阈值来判断是否发生运动，但是这样会比较依赖检测的结果精度。
就你的经验而言，你觉得我是否应该尝试一下这个repo里的一些专用的动态去除方法呢？

gt_cloud是如何得到的？

想问一下数据文件中SemanticKITTI数据集的gt_cloud是如何得到的？我暂时没有在您的项目中找到这一点

some question about gt

Hi Kin, thank you for your good work！I have a question about groudtruth, that is, how are static points and dynamic points generated in gt? Was it hand-marked？

Ask for Livox mid-360 dataset

cmake Error

Dear Dr. Zhang,

I am writing to express my sincere gratitude for your open-source contributions. I have encountered an issue during the cmake process and am seeking your assistance. The steps I have taken are as follows: cd methods/octomap && cmake -B build. The output was

Yours sincerely.

如何在线使用

非常感谢您的优秀工作，但我仍存在一些疑惑。
在其他平台看见您介绍erasor和removert是离线方案，octomap是在线方案，但未找到关于在线使用的介绍，请问如何在线使用，实时清除动态障碍物？

removert中的问题

博主你好，我最近也在研究removert方法，看了你的论文还有代码。目前有两个疑问：1.你的代码中将全局地图转化为rangimage的步骤没有将全局地图转化到当前扫描的局部坐标系下。2.我发现你的代码中和removert的原作者一样也没有revert部分，只有remove部分。
能解答一下为什么这样做吗，我最近在补充removert代码中的revert部分。还有就是在你的论文中描述了removert方法的缺点，有什么方法可以改进这些缺点吗。可以结合学习的方法先进行过滤在用removert吗？

KITTI data to pcd

Hello,
thank you for your open source work!
And I would like to know how you generate the pcd file of the adaptation code from the raw kitti data.

Is the GT pose relative to the lidar origin?

Thanks for your work. I have no experience using KITTI dataset. It seems the original odometry pose ground truth is for the camera. I wonder if the pose in the pcd is relative to lidar origin? It seems each pcd has been transformed with its pose in its VIEWPOINT field. Looking forward to your reply.

The demo in README cannot be run

cd methods/${methods_name} && cmake -B build && cmake --build build
./build/${methods_name}_run ${data_path, e.g. /data/00} ${config.yaml} -1