cann't get the accuracy in the paper

[Fridaycoder]

Hi, @fandulu I run the code you provide on SHREC dataset,but I cann't get the accuracy described in the paper,particularly,I got 90.1% on 14 gestures, and 75% on 28 gestures using DD-Net (filters=16),which is lower than the accuracy in the paper ,what's the problem?

[fandulu]

Hello, it seems the accuracies looks like

on 28 and

on 14 after I did a little fine-tuning.

[Fridaycoder]

@fandulu Thanks for replying. I got the approximate accuracy as the paper after fine-tuning, so the method benefit mostly from transfer learning？

[fandulu]

It is not a transfer learning since only the SHREC dataset is used. It is my fault to put a high learning rate and fewer epochs on the code. With a proper cyclic learning rate setting, you may even obtain a better performance than my paper claimed. Sorry for this problem.

[Fridaycoder]

@fandulu OK, I understand. Thank you for replying.

[chillcloud-dev]

@Fridaycoder
@fandulu
could you please share your adjustments where you could reach the accuracy of the paper ? The confusion matrix where values are higher than 95% can not be reached with the values provided with the code.
Adjusting the epochs (400) and learning rate (lr = 1e-2) didn't help much