search

cuishuhao / BNM

code of Towards Discriminability and Diversity: Batch Nuclear-norm Maximization under Label Insufficient Situations (CVPR2020 oral)
MIT License
265 stars 31 forks source link

cannot reproduce the results in the paper #1

Open d12306 opened 4 years ago

d12306 commented 4 years ago

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,

cuishuhao commented 4 years ago

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.

d12306 commented 4 years ago

@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!