Closed HustCK closed 3 years ago
Hi @CKkkkkkk
Thanks for posting question.
The code is same, but there are some modifications on the evaluation methods. In the official website it is based on 400/800 frames. I changed to larger value. Because if we only measure the 400 frames the result of loop closure is not significant.
Regards Han
Ok, I know, thanks for your answer very much!
Dear author, I have tried according to your answer. When I adjust from 400/800 to 2000, I can get 0.25% result on kitti sequence 00. Thanks for your answer again. By the way, do you know why many lidar odometry and SLAM system (eg. A-LOAM, Fast-LOAM, ISC-LOAM, LeGO-LOAM) could not work well on Kitti odometry sequence 04? The translational error of them are around 1.5-2.5%. Could you tell me the reason if you know?
Hi @CKkkkkkk
A possible answer is that the sequence 04 is too short, it only contains 270 frames.
Regards Han
请问,你们用的计算精度的工具是哪个吗
请问,你们用的计算精度的工具是哪个吗
You may write your own evaluation tools, or download from kitti website
OK ,thanks for your reply
我把lengths 设为 [800,900,1000,1100,1200,1300,1400,1500,1600,1700,1800,1900,2000] floam的旋转精度为0.5696, iscloam的旋转精度为0.4997% 这个和你设置的一样吗?
我把lengths 设为 [800,900,1000,1100,1200,1300,1400,1500,1600,1700,1800,1900,2000] floam的旋转精度为0.5696, iscloam的旋转精度为0.4997% 这个和你设置的一样吗?
In my case I write own evaluation code. I think the result is same when you set your length to infinity.
感谢,我学到了很多。
If there is not further question I shall close this issue
Dear authors, firstly I want to thanks for your code. I have tested your code on sequence 00 of Kitti odometry dataset, and the result of average translational error is around 1.17%, which is not match 0.24% published on your github page. The same problem occurs in Fast-Loam, where your result is 0.51% on sequence 00 and my result is 0.97%. Maybe I need to make some changes to the code base you released before testing on Kitti odometry? I would be very grateful if you could help me.