Thans for your excellent work, a few questions about the dataset and evaluation

[TongxingJin]

1. Do you have any plan to open-source the dataset in KAIST? Especially the 5 floors building shown in Figure 6;
2. How do you get the ground truth trajectory?

[dhchung]

1. It's hard to say. I want to upload it, but the dataset is too large (~ 200GB). I am currently finding some options to upload it.
2. I didn't. The trajectory is the result of NV-LIOM. You can see from the paper that it is just evaluated qualitatively.

[TongxingJin]

Thanks for your reply. Ok, waiting for your data. I think the staircase part shown in Figure 6 is decisive and enough for a simple test.

[TongxingJin]

For the loop closure part, you tried to project points from the candidate frame, as well as their normals, into current frame, and after dealing with visibility problem you compare the normal difference of the same pixel, right? And what do different colors mean in normal images?

[dhchung]

Thanks for your reply. Ok, waiting for your data. I think the staircase part shown in Figure 6 is decisive and enough for a simple test.

Ah yes! For the staircase dataset, I am planning to upload it.

For the loop closure part, you tried to project points from the candidate frame, as well as their normals, into current frame, and after dealing with visibility problem you compare the normal difference of the same pixel, right? And what do different colors mean in normal images?

Yes you are correct about the loop closure!

I forgot to turn off the imshow() in the code.. 😢 These images were just shown for me to debug the algorithm.
I converted the normal vectors into rgb in the images:
r = int((keyj.totalPoints->points[i].normal_x0.5+0.5) * 255);
g = int((keyj.totalPoints->points[i].normal_y0.5+0.5) * 255);
b = int((keyj.totalPoints->points[i].normal_z*0.5+0.5) * 255);
to see how the normal vectors are distributed in that scene.