BNM cannot reproduce the results in the paper

Hi, @cuishuhao , thanks for your code implementations, but when I tried to reproduce the results in the paper using the office-31 dataset, specifically from the dslr dataset to the amazon dataset, the final acc is around 69% (CDAN+BNM), can you try to reproduce the results using this version of code and release it here? are there any hyperparameters that you change in this transferring scenario?

Thanks,

Apr 04 '20 10:04 d12306

Hi, I ran eight experiments last day, but I still could not reproduce your results of D->A. I update the code for better reproducing. I tried on pytorch=1.0.1 and 1.3.1 and the results are similar to the paper. The iteration number is set to 5000 on D->W and W->D, and 10000 on others. While I do not think it matters. Now, the code could print the random number. If you could still obtain such awful results, please tell me the random number and environment.

Apr 05 '20 03:04 cuishuhao

@cuishuhao , thanks for your update, I guess it is caused by the initialization and the number of training steps, now I obtain an acc that is close to the one in the paper. Thanks, great work!

Apr 06 '20 01:04 d12306