search

erikwijmans / Pointnet2_PyTorch

PyTorch implementation of Pointnet2/Pointnet++
The Unlicense
1.53k stars 347 forks source link

furthest_point_sample from point2_utils is working differently from original pointnet++ #151

Closed tasbolat1 closed 3 years ago

tasbolat1 commented 3 years ago

Hi @erikwijmans ,

I have run a sample example on furthest_point_sample from point2_utils. The output are indices for xyz. This result differs from original fartherst_point_sample. Link for original tensorflow function is below: https://github.com/charlesq34/pointnet2/blob/42926632a3c33461aebfbee2d829098b30a23aaa/tf_ops/sampling/tf_sampling.py#L48

I would like to know how can we get exact results as in original. Thanks in advance

tasbolat1 commented 3 years ago

I have found the solution for this. It turns out that check for mag can be even smaller to give same results as tensorflow version. Have a look https://github.com/tasbolat1/Pointnet2_PyTorch/commit/b6229eedf934e7caa265efe4d0ce6f44f08446df

Closing for now