search

czq142857 / IM-NET-pytorch

PyTorch 1.2 implementation of IM-NET.
Other
196 stars 27 forks source link

inconsistency with AE training config and data #8

Closed yd-yin closed 3 years ago

yd-yin commented 3 years ago

https://github.com/czq142857/IM-NET-pytorch/blob/31b9cb2361321f03d5a99403533c0bee0dd92600/modelAE.py#L152-L163 The numbers here seems inconsistent with the data (sample 16 * 3 2 points for 32^3 voxels, 32 ** 3 points for 64^3 voxels), which may lead to incomplete training. https://github.com/czq142857/IM-NET-pytorch/blob/31b9cb2361321f03d5a99403533c0bee0dd92600/modelAE.py#L321

Also, can I simply assign self.load_point_batch_size as data_points.shape[1]?

czq142857 commented 3 years ago

The numbers are consistent with the data, if you downloaded the data from the correct repo. Please do not use the data or the point sampling code from other repos as they may be incompatible.

Yes, you can simply assign self.load_point_batch_size as data_points.shape[1].

yd-yin commented 3 years ago

That's true. Thanks a lot.