search

sitzikbs / Nesti-Net

Normal estimation for unstructured 3D point clouds using convolutional neural networks and a Mixture of Experts scale predictor.
Other
45 stars 13 forks source link

Question about training and testing noise level #3

Closed ziruiw-dev closed 4 years ago

ziruiw-dev commented 4 years ago

Hi @sitzikbs,

Thanks for releasing your code.

I am a little confused about what noise level do you use in training and testing. When you perform a training, did you train with all noise levels at once then test on different noise levels? or did you do (train on noise level A - test on noise level A) pairs?

Best, Ryan

sitzikbs commented 4 years ago

Hi @Ryan-Zirui-Wang , I used the same train-test split as in PCPNet. There is a .txt file containing the exact list of files for training and testing when you download the dataste. In my code, I put trainingset_temp.txt as a temporary file that contains much less models just so you can see that all is running well. If I am not mistaken, the original trainingset_whitenoise.txt file contains training examples with clean + 3 different noise levels (0.1, 0.05, 0.01) = total 32 files. Trainng is done on all of them together, not separate. The test is done separately on each of the noise levels. Best, Itzik

ziruiw-dev commented 4 years ago

Hi @sitzikbs, many thanks for your reply. Best, Ryan