test

[tang-y-q]

Hi,thanks for your good job!
I'd like to ask you some questions.
Can I reconstruct my data directly using your trained parameters?
When I use your pre-trained network to reconstruct my data, i get a bad result.
Is this because the network is not generalization capabilities enough?

[ThibaultGROUEIX]

Hi @tang-y-q

Can you share an example of your data please?
Are you following this script ? https://github.com/ThibaultGROUEIX/AtlasNet/blob/master/doc/demo.md#usage

Cheers,
Thibault

[tang-y-q]

Hi @ThibaultGROUEIX
thank you for your reply.
I rebuilt my data according to the method you said, but the effect was not ideal.
And here is my data.
data.zip
I sampled the point cloud on the grid model of turbine blade, and then reconstructed the surface of the point cloud.
However, there is a big gap between the surface I reconstructed and the original model.

best，
Yukon

[ThibaultGROUEIX]

Looking at your point cloud and the reconstruction, it's most likely because your point cloud is out of distribution of the ShapeNet dataset (that AtlasNet was trained on).