search

charlesq34 / pointnet

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
Other
4.74k stars 1.45k forks source link

The fluctuation of the performance during training #52

Closed maggie0106 closed 6 years ago

maggie0106 commented 6 years ago

Hi, Thanks for sharing your amazing work! I have a quick question. I have trained several trails of the point net on the ModelNet 40 dataset with all the default parameters in the code. I saw quite big performance fluctuation during the training, so it's a bit hard to pinpoint when the model converges. So is there any way to alleviate this performance fluctuation problem? Thanks

Sincerely, Maggie

charlesq34 commented 6 years ago

Hi Maggie,

The dataset is quite small (in some categories there are only 100 objects for training and a few for testing) so the result is not stable. What I usually do is to observe the loss curve and "early stop" the training at 200 or 250 epochs.

For stable evaluation I suggest you use the evaluation code which evaluates point cloud in all rotations instead of just the default rotation provided by ModelNet40.

Hope it helps

maggie0106 commented 6 years ago

Hi Charles, Thanks for replying me! I will try the method u mentioned.

Cheers, Maggie