Giters

rabbityl / lepard

[CVPR 2022, Oral] Learning Partial point cloud matching in Rigid and Deformable scenes

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Some question about the paper

Gilgamesh666666 opened this issue · comments

commented

Hi,

Thank you for sharing your excellent work! I have read the paper recently, and I have some questions about the position encoding. In your paper, you add position encoding to three places: self-attention, cross-attention, confidence matrix. And as you mention in the paper, position encoding in cross-attention and matching may mislead the registration at the beginning. So, I wonder if you have compared the performance at the situation you remove the position encoding at the cross-attention part and the confidence matrix part at the first stage but retain them at the second stage. Plus, I think the figures in your paper are very beautiful. So could you please tell me how you draw the point cloud and correspondences of figure 7 in your paper and the point cloud video in the readme? I will be very appreciative if you can give me some hints about the above question. Thank you very much!

commented

We did not check what happens if remove them, but agree it worth trying. For the drawing, we transfer all points/lines to meshes and render them with blender's cycle engine. the elephant animation is done via open3d.

commented

ntion part and the confidence matrix part at the first stage but retain them at the second stage. Plus, I think the figures in your paper are very beautiful. So could you please tell me how you draw the point cloud and correspondences of figure 7 in your paper and the point cloud video in the readme? I will be very appreciative if you can give me some hints about the above qu

Sir, It is an excellent idea ! Did you do this experiment? will the result increase or decrease?

commented

ntion part and the confidence matrix part at the first stage but retain them at the second stage. Plus, I think the figures in your paper are very beautiful. So could you please tell me how you draw the point cloud and correspondences of figure 7 in your paper and the point cloud video in the readme? I will be very appreciative if you can give me some hints about the above qu

Sir, It is an excellent idea ! Did you do this experiment? will the result increase or decrease?

Sorry, I didn't. I am a little busy these days :(