ktgiahieu / fast-ppf Goto Github PK

I am using your project now, but the background filtering part cannot be completely removed after inputting my data and scene point cloud. How should I adjust it, dear author?

Thanks to the author's excellent code, I got the correct result, but it is obvious that there are still four complete bearings in the figure that have not been identified. Why are relatively complete objects not identified?

Hi,

I tried to run the algorithm and found out that the result is different.
Any ideas on how to eliminate the misaligned part?

I have some problems with PCL. Can you upload the PCL source files and relevant third-parties files to the internet.thank you.

Hi author, thank you very much for your open source project, I am running the code under Window system, during the run, everything is working fine in step 5, however, when I run to step 6, only the output is available, the visualization graph exits automatically and the program terminates. For your code, I did not make any changes. If you have time, could you help to check what went wrong, thank you very much. The experimental platform I am using is QT Creator and PCL library version 1.8.1.
Step 5

Step 6 When I run to step 6, the visualization automatically exits and the code run terminates without running to step 7.

Thank you for your open-source contribution! During the process of learning and using your code, I achieved good test results with the provided data. However, when I applied my own model and data, the performance was not satisfactory. I attempted to increase the number of ICP iterations (2000 iterations) and the point cloud sampling ratio (0.02), but it did not improve the poor results. Do you have any suggestions?Additionally, I found that the sampling measurements in this project are not suitable for large-scale models (in the order of meters). Considering both large-scale models and small-scale scenes, does it seem impossible to modify this algorithm to make it applicable?

I can see that your code include some .h files of vtk. Dose this code depend on that lib?

Dear author, thank you for your selfless open-source contribution. While using your project, I encountered an issue with the 'leaf_size' parameter being too small during the filtering stage, which resulted in insufficient storage for voxel node data due to its 32-bit format. I would like to know if using Octree filtering would be a perfect solution to this problem. I prefer not to sacrifice accuracy to resolve this issue (typically, expanding from 0.001f to 10.0f), as I find this approach unacceptable.

Dear Author:
Fortunately, I have read your project and compiled it successfully. The code provided by you can also be tested, and the effect is better. However, when I tested with other data sets, the error "Leaf size is too small for the input dataset" appeared. Then, I use third-party software to reduce the model and field cloud scale of the dataset, and the reduced dataset is basically the same as the dataset you provided (due to the use of third-party software, I am not sure what proportion of the reduction is in this process).
When I run the program again, the error "Leaf size is too small for the input dataset" no longer appears, but the problem "Instance no Aligned" appears again. I didn't get an answer when I consulted the data, so I want to ask you. Is it because the data set model used for me is inconsistent with the reduction ratio of the field cloud?
Looking forward to your reply,thank you!

Hi, I can't compile the VTK part of the VTK when I'm translating PCL. How do you compile it.

I have solved it, but I used not the previous QT Creator experimental platform, replaced the experimental platform with VS2019 and PCL1.11, and finally I can realize the whole process, thanks again for your reply.

Originally posted by @Yfanmeng in #13 (comment)

bool
pcl::computePPFPairFeature (const Eigen::Vector4f &p1, const Eigen::Vector4f &n1,
const Eigen::Vector4f &p2, const Eigen::Vector4f &n2,
float &f1, float &f2, float &f3, float &f4)

When I run the second step of this project, the FPS in the lower left corner of the VTK visualization is usually around 2000, which will cause the project to crash, but it seems to be only around 50~60 in your video. On the issue of alignment, I have followed your steps to modify the material in the PCL library and recompile PCL, and then run the project and still cannot align.

您好，我是松山湖材料实验室的团队负责招聘的工程师，我在猎聘看了下您的简历，感觉您比较符合我们的职位要求和技术能力，我们主要技术栈要求是C++,QT，图像算法和三维处理算法，方便聊聊吗？我的电话是17327892390。能否烦请帮忙加我微信（电话微信同号），我们详细聊一下。谢谢

Hello，
Thanks for your great job!But i have some questions about it.I didn't change anything,but when i run the program to test the data you provided,i got bad results.it's not consistent with your test results.

The result you provided shows below.

I didn't run the program for a few days.But when i run it today,error occurs.

And then,i push the program again form your repository.But when i run it,i got the same error.

Hello, why do you take visual points from six perspectives in the code? I don't think the paper mentioned different perspectives. Is there any advantage in doing so? Thank you!