search

haoyu94 / RoITr

Rotation-Invariant Transformer for Point Cloud Matching
MIT License
97 stars 13 forks source link

How should the parameters of the training kitti dataset be set #12

Open k794102577 opened 4 months ago

k794102577 commented 4 months ago

How should the parameters of the training kitti dataset be set

Ruye-aa commented 3 months ago

Hello, I have also been trying to conduct experiments on the KITTI dataset using RoITr recently, and found that the results are very poor due to issues with the computation of the normal vectors, which may be unrelated to the computation parameters

k794102577 commented 2 weeks ago

Hello, I have also been trying to conduct experiments on the KITTI dataset using RoITr recently, and found that the results are very poor due to issues with the computation of the normal vectors, which may be unrelated to the computation parameters

Hello, have you settled this question now

CorgiBoyG commented 2 weeks ago

Hello, I have also been trying to conduct experiments on the KITTI dataset using RoITr recently, and found that the results are very poor due to issues with the computation of the normal vectors, which may be unrelated to the computation parameters

Hello, have you settled this question now

Yes, I recently tried to experiment the model on modelnet40 as well as kitti and found that I ended up with poor results with the original parameters not adjusted. For example, on modelnet40, I calculated the error between the estimated transform matrix and the real transform matrix, e.g. rre is more than 40. How can I mitigate this problem and can you provide me with some good suggestions? Looking forward to your reply, thanks!