search

hshustc / CVPR19_Incremental_Learning

Learning a Unified Classifier Incrementally via Rebalancing
184 stars 43 forks source link

Question about 'class_means' #5

Closed cxy1996 closed 4 years ago

cxy1996 commented 4 years ago

Hello, Thank you for your work.

cifar100-class-incremental/class_incremental_cosine_cifar100.py mentioned that the class_means[:,current_cl[iter_dico],0] is for iCaRL (mean of exemplars) and class_means[:,current_cl[iter_dico],1] is for NCM (theoretical mean).

But in cifar100-class-incremental/eval_cumul_acc.py, you use the class_means[:,current_cl[iter_dico],1] as the final result of your method, does it mean that you use all the historical exemplars to calculate the mean of the classes?

wonda commented 4 years ago

Hello, I think there is no problem. All the class means are calculated and saved for the evaluation of iCaRL and NCM classifier. But the evaluation of the "CNN acc" (the proposed method) does not use the class means.

cxy1996 commented 4 years ago

Yes, you're right. I misunderstood NCM.