forward function must have ground transform

[adizhol-str]

Hello,
Thank you for your work and publishing the source code!
Why did you choose to put the ground truth correspondences in the forward pass of the model. This makes evaluation on "just" point clouds very difficult.
Thank you

[Gs-Smit]

I also find this problem, especially when I use demo.py to do same test, I find when the gt is wrong, the estimated trans result is wrong and when gt is right, the result can be right. Is some code wrong?