search

zhulf0804 / GCNet

Leveraging Inlier Correspondences Proportion for Point Cloud Registration. https://arxiv.org/abs/2201.12094.
MIT License
103 stars 12 forks source link

re-training 3dmatch #15

Open x1y9 opened 7 months ago

x1y9 commented 7 months ago

I tried retraining 3dmatch database, and get following log for val after about 20 hours, seems loss and recall not improved significantly, is this right?

image

Thanks

Kris-Xu96 commented 4 months ago

dear bro i meet the error. can you solve it?

File "E:\Point_Cloud\GCNet-main\losses\loss.py", line 153, in forward inds = torch.norm(coords_src[coors[:, 0]].type(torch.long) - coords_tgt[coors[:, 1]].type(torch.long), dim=1) < (self.pos_radius - 0.001)
IndexError: tensors used as indices must be long, byte or bool tensors