could not reproduce classification results

[rabbiahassan]

Hello ! Your work is very interesting.Thanks for sharing. I trained and tested the classification network as per the instructions given by you on github page but results are far from what you have posted in the paper. I would really ppriciate if you can comment on it.

[tiangexiang]

Hi, 0.927 classification accuracy is really strange, the plain backbone (even DGCNN) performs much better than this. Can you please post your training log? That will help identify the problem. Thanks!

[rabbiahassan]

Hi, 0.927 classification accuracy is really strange, the plain backbone (even DGCNN) performs much better than this. Can you please post your training log? That will help identify the problem. Thanks!

Thanks a lot for the reply.I am attaching the training log generated by the network in run.log file. run.log

[tiangexiang]

I have re-evaluated the repo on two different machines, both got consistent results (0.9371 and 0.9363).

Can you please doublecheck there are not any changes to the original code?

If the code is indeed unmodified, can you upgrade/downgrade pytorch to 1.4.0 (which is used in my machines) and try again?

Please let me know if the problem still exists.

[rabbiahassan]

Thanks a lot for reply.I will test on an other machine and will let you know.For record I didnt modify code at all.