A question of the accuracy of transform

[DoMoCo]

Great work!!!

When I tried to replicate this work, I found that I consistently failed to obtain an accurate transformation whenever loop closure was detected. I noticed that the influence of the reference planes is significant;

When I select one reference plane, the accuracy of plane fitting greatly affect the final transformation by P3P. Could it be that I misunderstood something here?

By the way, I test it outdoor.

[nuo-1]

Great work!!!

When I tried to replicate this work, I found that I consistently failed to obtain an accurate transformation whenever loop closure was detected. I noticed that the influence of the reference planes is significant;

When I select one reference plane, the accuracy of plane fitting greatly affect the final transformation by P3P. Could it be that I misunderstood something here?

By the way, I test it outdoor.

Hi, I have also encountered similar problems. Do you have any discoveries, thanks.

[JixuanLee]

The same question. I run this program on my own dataset, and I detect the loop closure by BTCs. But the loop closure result (loop_transform) is not accurate. Such as, my origin pose-diff between the current frame and loop frame is ( 4, 4, 4) ( just for example, and just show(x, y, z)), but the result loop_transform maybe (0.01, 0.02, -0.03). I test it outdoor too. Thank you.