About the compared method 'Reconstructing Space'

[XLechter]

Hi, thanks for sharing this great work. Could you kindly share the implementation of 'Reconstructing Space' in your paper? I didn't find the offical codes for this paper.

Thank you so much!

[IdanAchituve]

Hi Wenxiao,
Thanks for the kind words!
I implemented this baseline myself. Unfortunately, the implementation is on a private repo. I may add it to this repo, but it won't be in the near future as I need to organize the code.
Given the code in my repo though implementing this baseline should not be difficult. Let me know if you run into trouble.

Cheers,
Idan

[Eksanf]

Hi, I try to implemented the RS but there are some points I not quite understand. Accoridng to the paper "Split the space to 3×3×3 equally sized voxels, shuffling them and, assign the network to reconstruct the original order." I split the point cloud to voxels and shuffling them as new pointcloud, Then I modify the reconstruction_loss try to reconstruct the original order of points, But the Loss I get is very large and model is not seem to work. Could you kindly share the implementation of 'Reconstructing Space' in your paper?

Thank you so much

[IdanAchituve]

Hi,
I am not sure why you are experiencing these issues. Unfortunately, I haven't got around to publishing the code for that baseline yet.
Perhaps you need to rescale the loss? Maybe there are other minor implementation details that you missed? When I implemented it, I followed the exact protocol presented in the paper and aided the official implementation.

Idan