slamwang / fastergicp Goto Github PK

View Code? Open in Web Editor NEW

100.0 100.0 15.0 75 KB

an improvement of fast_gicp

CMake 1.35% Cuda 25.82% C++ 72.79% Shell 0.04%

fastergicp's People

Contributors

Stargazers

Watchers

Forkers

kinggreat24 chengyushi1 mfkiwl mingjitianming grandzxw hanhan119 zczhangcheng wangwssh whuzs cdb0y511 gary-robotics hanmakaidao robot-mh hiyyg weixiaopassking

fastergicp's Issues

Hi, May I know where fasterGICP is better than the fastGICP?

Invalid (NaN, Inf) point coordinates given to nearestKSearch

I tried dropping fasterGICP in place as a replacement for fastGICP but unfortunately, running fast_gicp::FastVGICPCuda.align or fast_gicp::NDTCUDA.align gives me the following error:

localisation: /build/pcl-OilVEB/pcl-1.8.1+dfsg1/kdtree/include/pcl/kdtree/impl/kdtree_flann.hpp:136: int pcl::KdTreeFLANN<PointT, Dist>::nearestKSearch(const PointT&, int, std::vector<int>&, std::vector<float>&)
 const [with PointT = pcl::PointXYZ; Dist = flann::L2_Simple<float>]: Assertion `point_representation_->isValid (point) && "Invalid (NaN, Inf) point coordinates given to nearestKSearch!"' failed.

Running with fast_gicp:FastVGICP.align gives:

[pcl::KdTreeFLANN::setInputCloud] Invalid input!

The same scenario works with fastGICP and I double check that the input pointclouds are valid like:

     bool check_cloud_is_valid(const pcl::PointCloud<pcl::PointXYZ>::Ptr& cloud) {                                                                                                                                 
       bool valid = true;                                                                                                                                                                                          
       auto it = cloud->begin();                                                                                                                                                                                   
       for (; it != cloud->end(); ++it) {                                                                                                                                                                          
         if (!pcl::isFinite(*it)) {                                                                                                                                                                                
           printf("Found a point with inf!");                                                                                                                                                                      
           std::cout << it->x << "," << it->y << "," << it->z << std::endl;                                                                                                                                        
           valid = false;                                                                                                                                                                                          
         }                                                                                                                                                                                                         
       }                                                                                                                                                                                                           
       if (!cloud->is_dense) {                                                                                                                                                                                     
         valid = false;                                                                                                                                                                                            
         std::cout << "is dense fail" << std::endl;                                                                                                                                                                
       }                                                                                                                                                                                                           
       return valid;                                                                                                                                                                                               
     }

Do you have any ideas? I can see in the fast_gicp repo https://github.com/SMRT-AIST/fast_gicp there have been some bugfixes since this repo was forked - maybe these contain the fix?

Can it read the topic of rosbag in LEGO-LOAM or LIO-SAM?

I found that this code reading point cloud reads every. bin, so can it read the topic of rosbag? Can it be used in LEGO-LOAM or LIO-SAM? Can I ask you if it's convenient? my QQ is 1501182998

Could you give a simple explanation for the setting value?

Hi, @SLAMWang. Thanks for your great work.
I find it may be useful for our project.
But could you give a simple explanation for the setting value?

I think the modification is done on fast_gicp.hpp and fast_gicp_impl.hpp. (correct me if I was wrong)
I think a example for kitti is below:

fasterGICP/src/kitti.cpp

Line 151 in 756ec24

fast_gicp::FastGICP<pcl::PointXYZ, pcl::PointXYZ> gicp;

but I am a little confuse about the setting value here (I think these values are related to the two-fold filter, pls update some comments to help us):

fasterGICP/include/fast_gicp/gicp/fast_gicp.hpp

Line 42 in 756ec24

double first_matching_error_;
Another thing is confused me : what is the [imageProjection.cpp],[featureAssociation.cpp] for ?

Performance on the kitti test set

你好@SLAMWang. 非常感谢你有趣的工作！看到你的论文，被你方法精度惊艳到了，论文中报告的KITTI训练集精度接近0.2%的ATE，而目前SOTA大概只有0.45%左右，可以说有了极大地提升。不知道你的方法在KITTI测试集的精度如何？为什么没有在测试集上传结果呢，感觉很有希望排到榜首位置。另外，我按照你的readme操作，并没有成功运行KITTI数据集测试，不知道哪里出了问题？非常感谢，期待你的回复

Save point cloud

Hello!
I am a student learning about SLAM. Thank you for sharing such a wonderful tool.
I want to add
pcl::io::savePCDFile( "cloud.pcd", *map, true );
to mapping thread to save point clouds, but I can't find where it is.
If you don`t mind, may I ask where I should add the code??
I hope to hear you soon!!

P.S. I`m not familiar with GitHub (and English) yet, so sorry if my way using "Issues" were wrong.

Hi, may I ask how to install your project?

Your installation step:

git clone https://github.com/SMRT-AIST/fast_gicp --recursive
mkdir fast_gicp/build && cd fast_gicp/build
cmake .. -DCMAKE_BUILD_TYPE=Release

enable cuda-based implementations

cmake .. -DCMAKE_BUILD_TYPE=Release -DBUILD_VGICP_CUDA=ON

make -j8

seems not match this project.
The projected used is fast_gicp;

[pcl::KdTreeFLANN::setInputCloud] Invalid input!

unable to test the package

Method of evaluation

您好@SLAMWang ，请问您论文里面的误差评估是采用什么方法，比如：

请问是用这个kitti误差评估的方法吗，A. Geiger, P. Lenz, andR. Urtasun, “Are we ready for autonomous driving?the kitti vision benchmark suite,” in Proc. IEEEConf. Comput. Vis. PatternRecognit., 2012, pp. 3354–3361.

RGBD camera pointcloud registration

Hi, thanks for your great work.
Is fasterGICP suitable for pointcloud from RGBD camera?
I have tried your algorithm to registrate pointcloud generated from RGBD camera, but the result is not satisfactory. Could you give me some suggestions to improve the performance in RGBD camera's pointcloud?
thank you.