HPPLO-Net Train Problem
Opened this issue · 0 comments
yanboowang commented
Hello, thank you for your great work on HPPLO-Net. While training the network, I find that it seems that the pointcloud information is not used in model/voxelnetvlad/pointPWC/pointPWC.py and the icp loss cannot converge. Because in param/configvoxel.yaml, pwcnonlocal is set to be [null], thus feat1_l0 = self.level0(color1) .
But if I set pwcnonlocal: ["flow"]to use the pc1 and pc2 information, nonlocalnet will go wrong since idx1 and idx2 were set to None before in Flowodometry in model/voxelnetvlad/voxelnetvlad.py.
Could you please help me debug this or upload the correct code? Thank you!
